A reuse keeps what a person decided after the agent read it

[Maya-Kid]

Second of three PRs from my adoption plan on #725; the first merged as #1257.

Problem

0061's decision 3 says "A signature a person decided stays as the person left it", but not against what. On dev 040a533 a reuse (map_to) writes every shape and kind word it names as the adopter's binding. If a colleague binds one of them elsewhere after the agent read it, adopting overwrites that binding (for a shape, also writing a regression case for the proposal's property) and answers {id}. Nothing tells the adopter, and a reuse whose every item was decided that way is still adopted.

Change

• What the agent read. For a reuse, propose_locked stores the decided_at of each binding it read: seen_decided_at on each shape, and a kind_words_seen_decided_at map in signatures. merge_map_to keeps the first round's values.
• Keep rule. Adoption locks each binding row FOR UPDATE before reading it. A person's decision whose decided_at differs from the recorded read was made after it. The check is equality, not order: decided_at is the deciding transaction's start, so a decision begun before the read can commit after it. With no recorded read (a proposal saved before this change) every person's decision is kept, with no created_at fallback (Review queues are organised by what a decision changes #725's no-legacy rule).
• Agreeing and differing. A later decision with the same property and direction, or the same class, counts as written: it is not rewritten or listed, and a shape keeps its regression case. A differing one stays, gets no case, and is listed in skipped with the person's decision. A reuse with nothing left to write answers 409 proposal_stale and writes nothing.
• Response and workbench. POST /ontology/proposals/adopt answers {id, skipped}. When decisions were kept, the toast says how many; proposal_stale gets en and zh strings.
• Adopting a proposal commits the element, its follow-up work and the decision together #1257 review follow-ups. A class reuse queues align_phrases only when it wrote a kind word, alignment_busy gets en and zh strings, and api.decideProposal accepts only "rejected".
• Docs. A dated revision note under 0061's decision 3.

Verification

a_reuse_keeps_what_a_person_decided_after_the_agent_read_it runs two propose rounds against a scripted model, then adopts four reuses and one seeded without read times:

• a person's none made before the read is replaced, on a shape and on a kind word whose row the second round extended;
• later bindings elsewhere stay and are listed: a kind word's, and a shape's committed while the adoption waits for its row, its decided_at a day before the read;
• a later agreeing shape binding is not rewritten and keeps its case;
• a class reuse whose only kind word was later bound the same way answers 200, lists nothing and queues no align_phrases;
• the reuse whose only kind word was later bound elsewhere, and the seeded one over a person's binding, answer 409 and add no job or case.

On dev 040a533 with only this test, it fails at the first adoption (200, no skipped): dev rewrites all seven bindings, the colleague's too, and adopts every reuse.

Linux x86_64 container, Rust 1.98.1, PostgreSQL 16.15 + pgvector 0.8.6, Poppler 25.03, DB required (UTOPIA_TEST_REQUIRE_DB=1), node 20 for web, commit 54ac182 on dev 040a533: cargo fmt --all --check and clippy clean; cargo test --locked --workspace 1,576 passed, 0 failed, 6 ignored; web pnpm test 281 passed, pnpm build clean.

Limits

• No version check yet: a proposal refreshed after the person read it is adopted as it now stands.
• A binding row deleted since the agent read it (a class or property deletion cascades) cannot be locked; a person's new decision on it committing meanwhile can be overwritten.

[WaylandYang]

Thank you. Recording the decided_at the agent read and comparing by equality is right, and locking each binding row before reading it means a colleague's decision that commits during the adoption is kept and listed in skipped rather than overwritten; the test checks that by committing while the adoption waits. The #1257 follow-ups are all in. Two optional simplifications for later: build the skipped shape from the proposal's own shape rather than restating same_shape's four keys, so the regression-case filter can't drift from it; and now that alignment_busy is in the global error table, reviewErrors.ts could use that entry instead of keeping a second translation. Merging.