Add parallel PR review processing with queue depth monitoring

[Tyler-Rak]

Summary

Adds concurrent PR review processing capability with semaphore-based queue depth monitoring to handle multiple PRs efficiently while preventing resource exhaustion.

Features

1. Parallel Webhook Endpoint (/webhook-parallel)

• New endpoint for concurrent PR review processing
• Uses asyncio.create_task() for true parallel execution
• Semaphore-based concurrency limiting

2. Configurable Concurrency Control

[bitbucket_app]
enable_parallel_reviews = true
max_concurrent_reviews = 3

3. Queue Depth Monitoring

• Tracks active review tasks globally
• Logs queue depth metrics (queue_depth, active_reviews, review_start_time)
• Thread-safe task tracking with asyncio.Lock

4. Comment Filtering Enhancement

• Filters out non-actionable comments (UNAPPROVED, NEEDS_WORK)
• Only processes reasonable comments on parallel endpoint
• Prevents duplicate processing of approval/needs-work events

Changes

Core Implementation

pr_agent/servers/bitbucket_server_webhook.py:

• Added /webhook-parallel endpoint (200+ lines)
• Global review_semaphore for concurrency limiting
• Global active_review_tasks dictionary with async lock
• Queue monitoring with structured logging
• Comment filtering for parallel endpoint

tests/unittest/test_bitbucket_server_webhook.py:

• Added 174 lines of unit tests
• Tests for comment filtering logic
• Tests for queue monitoring

pyproject.toml:

• Version bump (tracking changes)

Logging Fix

• Changed log_context["event"] → log_context["webhook_event_type"]
• Fixes conflict with function parameter event= in logger

Benefits

1. Improved Throughput: Process multiple PRs concurrently
2. Resource Protection: Semaphore prevents overload
3. Observability: Queue depth metrics for monitoring
4. Reduced Noise: Filters non-actionable webhook events
5. Backward Compatible: Original /webhook endpoint unchanged

Configuration

Enable parallel processing in .secrets.toml:

[bitbucket_app]
enable_parallel_reviews = true
max_concurrent_reviews = 3  # Adjust based on resources

Testing

✅ Unit tests added: 174 lines of test coverage

• Comment filtering scenarios
• Queue monitoring logic

Deployment Notes

• Default behavior: Parallel reviews DISABLED by default
• Migration path: Can enable gradually per environment
• Monitoring: Check logs for queue_depth and active_reviews metrics
• Tuning: Adjust max_concurrent_reviews based on pod resources

Related Work

This PR works independently but pairs well with async subprocess improvements (#3) for even better performance on large PRs.

Documentation

See commit messages for detailed implementation notes:

• b3a2b338 - fix queue logging (queue monitoring implementation)
• 1f726954 - only filtering reasonable comments (comment filtering)