feat(git): declare the board and notes merge=union

[tieubao]

What

The repo now carries a .gitattributes declaring _meta/BACKLOG.md, _meta/backlog-staging.md and docs/implementation-notes/*.md merge=union. It never had one.

Board row: ID-886.

Why

The kit shipped every union mechanism (wrap's carry-across-pull, the union re-merge, board dedupe-all) and no declaration of its own, so git check-attr merge -- _meta/BACKLOG.md answered unspecified. Two sessions adding a row collided by hand: one sitting resolved the same stash-pop conflict four times, each time a stash push, an ff-only merge, a conflicted pop, a throwaway script over the hunks, and a reset to leave the row unstaged. ops-toolkit declares the same file union and never needed it.

The brief asked for something else

The brief asked for a hunk resolver on bin/wrap apply's POP CONFLICT path, on the premise that git stash pop leaves markers on a union-declared file. Refuted before any code:

1. A measured repro (upstream row and local uncommitted row in the same hunk) has the pop exit 0 with both rows, no markers, the file unstaged. Binary-declared variants do fail, and leave no markers, so a resolver never sees them.
2. tests/test-wrap.sh has asserted this since the pull-past-dirty work, and lib/wrap/wrap.sh states it in the _unstash comment.
3. The pop conflicted because the file was not declared union HERE, which the proposed check-attr == union gate would not have fired on either.

Evidence and the rejected alternative: docs/implementation-notes/gitattributes-union.md.

The one code change the declaration forced

Declaring the board union makes wrap's carry-across-pull path reachable for it. That path inserted below the first --- line; the board has none, so a carried row would have landed at line 1, above the title. _union_carry_back now keeps the anchor rule for a file that has an anchor and runs git merge-file --union over pulled/base/local for a file that has none. Narrowing it that way leaves every asserted contract intact: the driver orders the incoming entry above the local one, which is backwards for a newest-first LAB_LOG.

A first attempt placed the block after its local neighbour. Review broke that four ways (blank neighbour, repeated neighbour, non-contiguous lines, empty first line); the driver answers all four in less code.

Verification

Check	Result
bash tests/test-gitattributes-union.sh	27/27, exit 0
bash tests/test-wrap.sh	all 508 passed, exit 0
bash tests/test-meta.sh	853/853, exit 0
bash lib/gate/negctl.sh . "bash tests/test-gitattributes-union.sh" "rm .gitattributes"	PASS (green, red under mutation, green restored)
Review	code-reviewer on the bash diff, six findings, four fixed by the driver, one an accepted documented tradeoff, one stale

Proof of done: docs/verification/gitattributes-union.md.

Accepted tradeoff

Union never drops a line, so two branches flipping the same row's Status yield that row twice instead of a conflict. The next board set then refuses with "matches N rows; dedupe first", which is loud and recoverable. The .gitattributes comment names it and says to check which side's Status dedupe-all kept. ops-toolkit made the same call.