Fix 1.1: Make ResponseApplyPatchToolCall import conditional with try/except
for compatibility with openai==1.100.1 (CI environment)
Fix 1.2: Move Router creation inside mock context in vector store tests
so mocks are applied before Router captures function references
Fix 1.3: Update test_model_group_info_e2e to check for 'anthropic/*'
wildcard group instead of specific model names not in proxy config
Fix 2.1: Increase redis cache test sleep from 1s to 5s
Fix 2.2: Increase spend accuracy test sleep from 25s to 45s
Fix 2.3: Add 0.5s sleep between budget test calls
Fix 2.4: Increase vertex AI spend test sleep from 20s to 40s
Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>