Bound lock waits for human phrase alignment review decisions (#798, #800)

[rollroyces]

/materials requests review requests — bounded lock waits (#798, #800)

Survey follow-up to #798's closing question — "every synchronous handler that takes a lock a background job can hold".

Handler	Outcome
decide_alignment_kind_word	Fixed in #828 by decide_and_apply_human. Closed.
decide_alignment_phrase	This PR. Calls materialize inline at review_routes.rs:1280. The worker align_phrases holds the same advisory lock, and the wait has no cap. #800 was closed as completed when design landed on #841 — but the code fix never followed. This PR does that.
decide (generic review)	Calls merge_entities → temporal.rs per-timeline advisory locks. Real but invasive (multiple lock sites). Deferred.
decide_mapping, decide_defect, decide_violation, decide_pending, decide_proposal, apply_import	Single-row UPDATE or no relevant lock; deferred.

What this PR does

• materialize_human (utopia-store) — same body as materialize, but sets SET LOCAL lock_timeout = '2s' at the top of the transaction and maps PostgreSQL 55P03 to AppError::CodedConflict { code: "alignment_busy" } with a "please try again" message. Same pattern as #828 did for decide_and_apply_human.
• materialize (worker path) — untouched. Workers don't have a spinner to look at; their waiting policy is unchanged.
• decide_alignment_phrase — switched to materialize_human.

Tests

crates/utopia-store/tests/human_materialization_has_a_lock_budget.rs:

• With the typed_materialize lock held by another connection, materialize_human returns CodedConflict("alignment_busy") well within 10 seconds, with a retry message. Once the gate is released, a worker materialize completes normally and produces the expected Outcome.**
• With no contention, materialize_human matches materialize. SHOW lock_timeout returns empty after the call, confirming SET LOCAL did not leak to the session.

Validation

• cargo check --package utopia-store clean
• cargo check --package utopia-server clean
• cargo clippy --package utopia-store --all-targets -- -D warnings clean
• cargo clippy --package utopia-server --all-targets -- -D warnings clean
• cargo fmt --check clean

Why this and not the bigger cut

decide / merge_entities would catch the same bug pattern at multiple temporal.rs lock sites. Doing it cleanly is a second PR's worth of test surface. The two materialize_human regressions above pin the phrase path; the merge path can land as a follow-up with its own survey note. If you'd rather do both at once, say so and I'll extend the cut — but I'd rather ship the smaller reviewable PR first.