fix(lint): make the rebased branch pass the delta-vs-base lint gates

The rebase side failed CI at the first lint step, so the later gates had
never run; running them locally surfaced three more. All four are
mechanical:

1. ruff I001: the rebase resolution left the anthropic transformation
   import groups different from upstream's layout; restore upstream's own
   grouping (third-party block, then litellm).
2. ruff-strict B010: the usage change replaced one setattr with two, so
   the file counted one more B010 than the merge-base.
3. type-discipline LIT011 / basedpyright reportAttributeAccessIssue:
   writing `model_response.usage = ...` is an in-place store on a
   parameter; the two stores added 2 LIT011 and 2 attribute-access
   diagnostics over the merge-base ceilings.
4. basedpyright unknown-type rules: reading
   `completion_response.get("usage")` into a `Final` local added
   reportUnknownMemberType and reportUnknownVariableType. A Final name
   takes its inferred type, so an annotation cannot widen it away; the
   value has to stop flowing out of an unknown dict into a new local.

Resolved as one shape rather than four patches: compute the value with a
conditional expression and write it back through the single `setattr`
upstream already used on that line, so the B010, LIT011, attribute-access
and basedpyright rule counts all stay at their merge-base totals, with no
suppression comment anywhere.

No behavior change: a missing key and an explicit null both still mean
"no usage reported" (asserted by
test_transform_parsed_response_without_usage, both variants).

Signed-off-by: apex-mochen <2756823972@qq.com>

litellm/llms/anthropic/chat/transformation.py

@@ -6,6 +6,9 @@ from types import MappingProxyType
 from typing import TYPE_CHECKING, Any, ClassVar, Final, NoReturn, cast
 
 import httpx
+from pydantic import BaseModel, ValidationError
+from typing_extensions import ReadOnly, TypedDict
+
 import litellm
 from litellm.constants import (
     ANTHROPIC_MIN_THINKING_BUDGET_TOKENS,
@@ -92,8 +95,6 @@ from litellm.utils import (
     supports_reasoning,
     token_counter,
 )
-from pydantic import BaseModel, ValidationError
-from typing_extensions import ReadOnly, TypedDict
 
 from ..common_utils import (
     AnthropicError,
@@ -2739,26 +2740,24 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
             map_finish_reason(completion_response["stop_reason"]),
         )
 
-        usage_object: Final = completion_response.get("usage")
-        if usage_object is None:
+        usage: Final = (
+            None
+            if "usage" not in completion_response or completion_response["usage"] is None
+            else self.calculate_usage(
+                usage_object=completion_response["usage"],
+                reasoning_content=reasoning_content,
+                completion_response=completion_response,
+                speed=speed,
+            )
+        )
+        if usage is None:
             # #44535: keep usage absent (None) instead of fabricating a 0/0
             # usage object. Spend tracking and budgets must be able to tell
             # "no usage reported" apart from a genuine zero, otherwise a
             # response that omits usage would be recorded as free even though
             # the upstream consumed tokens.
-            setattr(model_response, "usage", None)
             _hidden_params["usage_missing"] = True
-        else:
-            setattr(
-                model_response,
-                "usage",
-                self.calculate_usage(
-                    usage_object=usage_object,
-                    reasoning_content=reasoning_content,
-                    completion_response=completion_response,
-                    speed=speed,
-                ),
-            )
+        setattr(model_response, "usage", usage)
 
         model_response.created = int(time.time())
         model_response.model = completion_response["model"]