[DanielNetzer]

Implements issue #242.

Molecule: mol-242-8d304c351bc4ca45bdf6fcb8
Formula: forgecell-github-issue
Base: 6bd19d73efc11cac4400d97256a320aecaacae0b
Delivery: d1fad87ec4ccce85ddea535626eace6f541c05a2

Awaiting exact-commit CI and human review. Harness-reported checks are not independent CI evidence. No merge, deployment or issue closure is authorized.

[DanielNetzer]

Adversarial review: remaining LOW findings (not fixed in this slice)

1. The cache ignores new numbers within the TTL. A just-published PR, or a new Molecule on a closed Issue, shows "not checked" for up to 60 s. The UI flickers but never shows a wrong state. Fix: treat the snapshot as not fresh when the wanted set has numbers it lacks.
2. TestFetcherByteCapWithTheRealRunner spawns sh -c head -c …. That's POSIX-only, which is fine while CI runs on Linux and macOS.

Fixed in this slice:

• row 15 timing (the reopen must come after the merge);
• Startable fails closed on stale data;
• a panic in the fetcher no longer freezes the cache, and the single-flight waits are bounded;
• Unicode format characters (bidi overrides) are stripped from titles;
• the missing matrix test cases are added.

[DanielNetzer]

Merged main into this branch (560c62f) to resolve a server.go conflict with #268 (#251) and #267 (#254): union of the handler fields (forge sync + amend/numstat/diffs), keeping runIssue: molecule.RunIntake. Verified on the merged tree: build, vet, go test -count=1 ./internal/ui/... ./internal/forgesync/... ./internal/delivery/... pass.