Migrate from Chat Completions to Responses API

[pamelafox]

Summary

This PR migrates the OpenAI API usage from the legacy Chat Completions API to the unified Responses API.

Changes

Client Migration

• Import: AsyncAzureOpenAI → AsyncOpenAI
• Endpoint: Uses base_url with /openai/v1 path instead of deprecated azure_endpoint
• Auth: azure_ad_token_provider → api_key (token provider still works)
• Removed: api_version parameter (no longer needed for Responses API)

API Call Migration

• Method: chat.completions.create() → responses.create()
• Input format: messages → input with structured content items
  • User/system messages use type: "input_text"
  • Assistant messages use type: "output_text"
• Added: store=False to avoid server-side state retention

Streaming Events

• Before: ChatCompletionChunk with choices[0].delta.content
• After: response.output_text.delta and response.completed events

Infrastructure Cleanup

Removed AZURE_OPENAI_API_VERSION from:

• infra/aca.bicep
• infra/main.bicep
• infra/main.parameters.json
• .github/workflows/azure-dev.yaml

Dependencies

• Updated openai>=1.108.1 in pyproject.toml
• Regenerated requirements.txt (now uses openai 2.16.0)

Tests

• Simplified test mocks for Responses API streaming format
• Updated snapshot files to match new response structure
• All 4 tests passing ✅

Files Changed (11)

File	Description
src/quartapp/chat.py	Main migration - client init and API call
src/pyproject.toml	Updated openai version constraint
src/requirements.txt	Regenerated dependencies
tests/conftest.py	Simplified mock fixtures
tests/test_app.py	Updated managed identity test
tests/snapshots/*	Updated 2 snapshot files
infra/aca.bicep	Removed API version env var
infra/main.bicep	Removed API version param/output
infra/main.parameters.json	Removed API version parameter
.github/workflows/azure-dev.yaml	Removed API version env var

Testing

• All unit tests pass
• Ruff linting passes
• Manual testing with Azure OpenAI deployment (recommended before merge)