feat: add per-request DB query tracing with ContextVar

[ahmedelgyar1]

Summary

Added opt-in query tracing to the metadata service. When QUERY_TRACING_ENABLED=1 is set, it logs a single summary line per HTTP request showing the total DB queries, rows returned, and execution time.

Context

Fixes #5
Right now, we can't tell how many database queries are triggered by a single API request or how long they take. This makes it hard to debug performance issues, especially when a single client operation fans out into many HTTP requests. This PR adds server-side observability to measure the exact DB cost of any request.

Screenshots

Changes Made

• Added a new query_tracing module with an aiohttp middleware.
• Used contextvars.ContextVar to keep track of queries per request. This prevents logs from getting interleaved when there are many concurrent requests.
• Added timing and row counting logic inside execute_sql().
• Added QUERY_TRACING_ENABLED env variable (defaults to 0 / off).
• Zero overhead when disabled: If the env var is not set, the tracing logic is completely bypassed.

Testing

• Added unit tests in query_trace_test.py to verify:
  • Metrics are aggregated correctly per request.
  • SQL strings are truncated to 200 chars to avoid log spam.
  • Tracing context is properly cleaned up after the request finishes.
  • The feature is a complete no-op when disabled.
• Manually tested locally via Docker: Ran QUERY_TRACING_ENABLED=1 docker-compose up and hit /flows and /runs endpoints using the Swagger UI to verify the logs appear correctly in stdout.