feat(metrics): HTTP metrics middleware with per-minute rollup table

[anoop-narang]

Summary

• Adds a metrics_request_rollup_minute table (v7 migration, SQLite + Postgres) for fast per-minute request/error counts and latency aggregates
• Introduces a src/metrics module with MetricsEvent, RollupBucket, and a background MetricsWorker that accumulates events in memory and flushes to the DB every 30 s via additive UPSERT
• Adds an Axum metrics_middleware (registered in the Tower stack) that captures route template, method, status code, and wall-clock latency per request — using try_send so the hot path is never blocked
• Wires the channel and worker into RuntimeEngine with clean shutdown support (final flush on drop)

Multi-instance safe: the UPSERT is additive, so multiple runtimedb instances sharing one catalog DB sum their counts correctly.
No double-counting: the in-memory bucket map is cleared after every flush.
p50/p99 percentiles for individual queries are already available via query_runs.execution_time_ms; the rollup table covers mean/min/max across all endpoints.

Test plan

• cargo test --lib — all 404 unit tests pass (includes new worker accumulation/bucket/shutdown tests)
• cargo test --test result_persistence_tests — 22 integration tests pass
• cargo clippy and cargo fmt --check — clean
• Manual: run server, fire requests, wait ~30 s, query SELECT * FROM metrics_request_rollup_minute ORDER BY minute DESC to verify rows populate with correct counts and latency fields