fix(ci): remove duplicate spec_path in prisma schema + address review feedback

- Remove duplicate spec_path field in LiteLLM_MCPServerTable (lines 280 and 308)
  that was causing all GitHub Actions matrix tests to fail with Prisma validation error
- Narrow exception assertion to only MidStreamFallbackError (Greptile review)
- Add upper bound to openai version constraint: >=2.8.0,<3.0.0 (Greptile review)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

docker/build_from_pip/requirements.txt

@@ -2,5 +2,5 @@ litellm[proxy]==1.82.1 # Specify the litellm version you want to use
 prometheus_client
 langfuse
 prisma
-openai>=2.8.0
+openai>=2.8.0,<3.0.0
 ddtrace==2.19.0 # for advanced DD tracing / profiling

litellm/proxy/schema.prisma

@@ -305,7 +305,6 @@ model LiteLLM_MCPServerTable {
   registration_url   String?
   allow_all_keys Boolean @default(false)
   available_on_public_internet Boolean @default(true)
-  spec_path             String?
   is_byok               Boolean  @default(false)
   byok_description      String[] @default([])
   byok_api_key_help_url String?

tests/local_testing/test_streaming.py

@@ -3114,9 +3114,8 @@ def test_unit_test_custom_stream_wrapper_repeating_chunk(
     if (loop_amount > litellm.REPEATED_STREAMING_CHUNK_LIMIT) and expected_chunk_fail:
         from litellm.exceptions import MidStreamFallbackError
 
-        with pytest.raises(
-            (litellm.InternalServerError, MidStreamFallbackError)
-        ):
+        # MidStreamFallbackError wraps InternalServerError when repeated chunks detected
+        with pytest.raises(MidStreamFallbackError):
             for chunk in response:
                 continue
     else: