refactor(tests): move reasoning_effort grid suite under llm_translation, drop v4 naming

- Drop the "v4" suffix throughout: it referred to the QA sweep iteration,
  not this test suite. There's only one regression suite, so just call it
  reasoning_effort_grid.
- Move tests/test_litellm/reasoning_effort_grid_v4/ -> tests/llm_translation/
  reasoning_effort_grid/. Two reasons:
    1. The parent tests/test_litellm/conftest.py installs an autouse fixture
       (isolate_host_aws_config) that clears every AWS_* env var before each
       test, which would silently skip every Bedrock cell.
    2. tests/llm_translation/conftest.py already wires up the Redis-backed
       VCR persister and auto-applies @pytest.mark.vcr to every collected
       item via apply_vcr_auto_marker_to_items. Living under that conftest
       means the suite gets cassette replay for free -- first CI run with
       provider creds records 231 cassettes, every subsequent run replays
       them with no live spend.
- Trim the suite's own conftest down to just the wire_capture fixture; the
  inherited llm_translation conftest covers the VCR plumbing.
- Drop the dedicated reasoning_effort_grid_v4_e2e CircleCI job. The existing
  llm_translation_testing job globs tests/llm_translation/**/test_*.py, so
  the suite is gated by an existing job with no new wiring.

.circleci/config.yml

@@ -576,48 +576,6 @@ jobs:
           no_output_timeout: 15m
 
       # Store test results
-      - store_test_results:
-          path: test-results
-  reasoning_effort_grid_v4_e2e:
-    docker:
-      - *python312_image
-    working_directory: ~/project
-    resource_class: large
-
-    steps:
-      - checkout
-      - setup_google_dns
-      - install_uv
-      - restore_cache:
-          keys:
-            - v1-uv-cache-{{ checksum "uv.lock" }}
-      - run:
-          name: Install Dependencies
-          command: |
-            uv sync --frozen --all-groups --all-extras --python 3.12
-      - save_cache:
-          paths:
-            - ~/.cache/uv
-          key: v1-uv-cache-{{ checksum "uv.lock" }}
-      # Grid v4 exercises reasoning_effort mapping against real Anthropic,
-      # Azure AI Foundry, Vertex AI, Bedrock Converse, and Bedrock Invoke
-      # endpoints. Per-route cells pytest-skip themselves when the matching
-      # provider env vars are absent, so PRs without credentials no-op.
-      - run:
-          name: Run reasoning_effort grid v4 e2e suite
-          command: |
-            mkdir -p test-results
-            uv run --no-sync python -m pytest \
-              tests/test_litellm/reasoning_effort_grid_v4/ \
-              -v \
-              --junitxml=test-results/junit.xml \
-              --durations=20 \
-              -n 4 \
-              --timeout=180 --timeout_method=thread \
-              --retries 2 --retry-delay 5 \
-              --max-worker-restart=5
-          no_output_timeout: 20m
-
       - store_test_results:
           path: test-results
   realtime_translation_testing:
@@ -2661,8 +2619,6 @@ workflows:
           filters: *main_branches
       - llm_translation_testing:
           filters: *main_branches
-      - reasoning_effort_grid_v4_e2e:
-          filters: *main_branches
       - realtime_translation_testing:
           filters: *main_branches
       - agent_testing:

tests/llm_translation/reasoning_effort_grid/conftest.py

@@ -1,6 +1,12 @@
-"""Shared fixtures for the reasoning_effort grid v4 e2e suite."""
+"""Shared fixtures for the reasoning_effort grid e2e suite.
+
+VCR wiring (Redis-backed cassette persister, auto-application of
+``@pytest.mark.vcr`` to every collected item, cassette-cache health summary)
+is inherited from ``tests/llm_translation/conftest.py``. This file only
+contributes the ``wire_capture`` fixture, which records the wire body
+LiteLLM sends upstream so each cell can inspect it.
+"""
 
-import os
 from typing import Any, Dict, List, Optional
 
 import pytest
@@ -12,10 +18,10 @@ from litellm.integrations.custom_logger import CustomLogger
 class _WireBodyCapture(CustomLogger):
     """Pre-call hook that records the outgoing wire body LiteLLM sends upstream.
 
-    `complete_input_dict` is the fully transformed provider request as set by
-    every provider transformation in `litellm/llms/**`. Capturing it here means
-    a regression anywhere in the transformation chain (strip, rename, drop)
-    surfaces as an assertion failure on the cell that depends on it.
+    ``complete_input_dict`` is the fully transformed provider request as set
+    by every provider transformation in ``litellm/llms/**``. Capturing it here
+    means a regression anywhere in the transformation chain (strip, rename,
+    drop) surfaces as an assertion failure on the cell that depends on it.
     """
 
     def __init__(self) -> None:
@@ -37,9 +43,6 @@ class _WireBodyCapture(CustomLogger):
     def latest(self) -> Optional[Dict[str, Any]]:
         return self.records[-1] if self.records else None
 
-    def reset(self) -> None:
-        self.records.clear()
-
 
 @pytest.fixture()
 def wire_capture():
@@ -50,12 +53,3 @@ def wire_capture():
         yield capture
     finally:
         litellm.callbacks = previous
-
-
-@pytest.fixture(scope="session")
-def vertex_credentials_path() -> Optional[str]:
-    """Resolve a usable Vertex credentials file path or None."""
-    path = os.environ.get("GOOGLE_APPLICATION_CREDENTIALS")
-    if path and os.path.exists(path):
-        return path
-    return None

tests/llm_translation/reasoning_effort_grid/grid_spec.py

@@ -1,12 +1,13 @@
 """
-Canonical post-fix expectations for the reasoning_effort grid v4 sweep.
+Canonical post-fix expectations for the reasoning_effort grid sweep.
 
-The QA sweep on https://github.com/BerriAI/litellm/pull/27039#issuecomment-4363363610
-covered 21 (provider x model) combos x 11 effort values (231 cells). The follow-up
-PR https://github.com/BerriAI/litellm/pull/27074 closed nine bugs surfaced by that
-sweep. This module encodes the post-fix expectations as a small rule set keyed by
-(model_mode, effort) and per-model capability overrides, then expands them across
-the model x effort matrix per route.
+The original QA sweep on
+https://github.com/BerriAI/litellm/pull/27039#issuecomment-4363363610
+covered 21 (provider x model) combos x 11 effort values (231 cells). The
+follow-up PR https://github.com/BerriAI/litellm/pull/27074 closed nine bugs
+surfaced by that sweep. This module encodes the post-fix expectations as a
+small rule set keyed by (model_mode, effort) and per-model capability
+overrides, then expands them across the model x effort matrix per route.
 """
 
 from dataclasses import dataclass, field

tests/llm_translation/reasoning_effort_grid/test_reasoning_effort_grid.py

@@ -1,6 +1,6 @@
 """
-End-to-end grid v4 regression suite for reasoning_effort mapping across
-Anthropic-backed routes.
+End-to-end regression suite for reasoning_effort mapping across the
+Anthropic-backed routes covered by the original QA sweep.
 
 Encodes the 21 (provider x model) x 11 effort matrix (231 cells) from the
 QA sweep on https://github.com/BerriAI/litellm/pull/27039#issuecomment-4363363610
@@ -13,9 +13,12 @@ against. Each cell asserts:
   - Status code returned by LiteLLM (200 vs BadRequestError -> 400) -- the
     regression signal for clean-error vs leaked-500 mappings.
 
-Hits real provider endpoints. Each route is skipped at runtime when its
-required env vars are absent, so PR builds without provider credentials no-op
-gracefully.
+Calls go to real provider endpoints, but the parent
+``tests/llm_translation/conftest.py`` auto-applies ``@pytest.mark.vcr`` to
+every collected item, so first run records cassettes (Redis-backed) and
+subsequent CI runs replay them with no live spend. Each route still skips at
+runtime when its required env vars are absent, so PR builds without provider
+credentials no-op gracefully.
 """
 
 import os
@@ -180,7 +183,7 @@ async def _call_messages(
 @pytest.mark.parametrize(
     ("route_name", "model", "effort", "cell"), _PARAMS, ids=_PARAM_IDS
 )
-async def test_reasoning_effort_grid_v4(
+async def test_reasoning_effort_grid(
     route_name: str,
     model: ModelEntry,
     effort: str,
@@ -209,7 +212,7 @@ async def test_reasoning_effort_grid_v4(
         raise
 
 
-def test_grid_v4_cell_count() -> None:
+def test_grid_cell_count() -> None:
     """Guard against accidental drops or duplicates in the grid spec."""
     assert len(_PARAMS) == 21 * 11, (
         f"expected 231 cells (21 provider x model combos x 11 efforts), "
@@ -217,7 +220,7 @@ def test_grid_v4_cell_count() -> None:
     )
 
 
-def test_grid_v4_route_coverage() -> None:
+def test_grid_route_coverage() -> None:
     """The grid must cover every route the original QA sweep covered."""
     route_names = {route.name for route in ROUTES}
     assert route_names == {