chore(deps): lock sqlalchemy 2.0.55

[lesnik512]

Locks SQLAlchemy at 2.0.55, released 2026-10-16. The sqlalchemy[asyncio]<2.1 pin stays until the OTel SQLAlchemy instrumentor supports 2.1. Its latest release, 0.66b1, still declares sqlalchemy<2.1.0.

2.0.55 fixes ticket 13673. When a task was cancelled a second time while the pool was closing an asyncio connection that the first cancellation had invalidated, the closed connection stayed in the pool. The next checkout got it and failed on first use. AnyIO cancel scopes hit this every time, because they keep delivering the cancellation until the task leaves the scope. With pool_pre_ping off (ADR 0012), only postgres_retry would catch that failure. The release also fixes related cancellation handling in the pool and in async context managers (tickets 12710, 13622, 13678).

Only sqlalchemy moves in uv.lock.

just test: 140 passed, 100% coverage. just test-migrations: 4 passed. The container runs SQLAlchemy 2.0.55.