A push to main recombines every open kept-fork pull, so its combined preview is never stale for long #9931

For: a kept fork's pull shows what would land now, not what would have landed before main moved. docs/features/upstream-landing.md says "A push to main alone does not recombine: the card marks itself stale … until the next push to the fork or the landing recombines."

Done looks like:

• [ ] After a push to a repository's default branch, every open pull from a kept fork into it gets combined/N rebuilt (src/git/combined.ts) on the new tip, off the request path, at most one at a time per repository, coalescing bursts of pushes.
• [ ] The card's stale line (src/core/combined_view.ts) clears once the rebuild finishes. A rebuild that fails leaves the card stale with the failure's sentence.
• [ ] test/combined-fork.test.ts (or a new test/combined-recombine.test.ts) pushes to main alone and asserts that combined/N now has the new main as a parent and that the card is no longer stale. It fails on main.
• [ ] docs/features/upstream-landing.md: the 'stale until next push' paragraph and the matching 'Not yet' item are updated.

Files: the push handler that already recombines on a fork push, src/git/combined.ts, src/core/combined_view.ts, a test, docs/features/upstream-landing.md.

Parked by the crew: the check stayed red through 2 automatic tries (pull #9941 closed). Planner: split this smaller, or rewrite what done looks like; a person can also say what to try.

Parked by the crew: the check stayed red through 2 automatic tries (pull #9976 closed).

Parked twice since a person last spoke here, so it is the planner's now. Every reason:

1. the check stayed red through 2 automatic tries (pull #9941 closed)
2. the check stayed red through 2 automatic tries (pull #9976 closed)

The planner splits it, replaces it with a fresh issue, or defers it; a person can also say what to try. Nothing starts here until then, and the board lists it.

Split by the planner: #9994 (the recombine pass on a push to main, wired from post-receive, with its test; the pull must name any red test) and #9995 (the failure sentence on the card, and the docs). Pulls #9941 and #9976 went red twice for a reason nobody found, so the work now lands in two smaller pulls.