Avoid partition DDL on relay startup

[bradseiler]

Why

Relay pods currently run partition DDL on every startup even when every required monthly partition already exists. CREATE TABLE ... PARTITION OF takes an ACCESS EXCLUSIVE lock on the hot parent table, so routine pod churn can unnecessarily queue writes behind startup DDL.

What

• Compute the expected monthly partition specifications up front and query PostgreSQL catalogs once for all expected partitions.
• Return immediately when every partition exists, avoiding parent-table DDL locks on the normal startup path.
• Run DDL only for missing partitions, retaining a second catalog check for races.
• Apply a transaction-local 2-second lock_timeout on the same connection as the partition DDL.
• Verify partition schema, parent, and range bounds before considering a partition ensured, including concurrent-creation and IF NOT EXISTS collision paths.
• Add unit and live PostgreSQL coverage for the lock-free fast path, subset creation, concurrent creation, bounded lock timeout/recovery, and unsafe same-name collisions.

Validation

• cargo test -p buzz-db partition::tests -- --nocapture
• cargo test -p buzz-db partition::tests -- --ignored --nocapture
• cargo test -p buzz-db
• just fmt-check
• just desktop-tauri-fmt-check
• git diff --check origin/main...HEAD
• Push hooks: branch-skew, file-size-check, rust-tests, desktop-tauri-checks

Generated with bb-expert