Review follow-ups: a path before a full stop, the write-time hold check, the public commit, the idle test #881

Closes #867

Four small follow-ups from the reviews of #806 and #833, one commit each; based on main with the newest main merged in.

• #867 (paths.ts): a directory that ends a sentence was left whole, so the forge's internal path showed in a run log. A full stop followed by whitespace or the end of the line is a boundary: wrote to <dir>. reads wrote to .; a sibling whose name goes on after the dot (<dir>.bak) is still left whole. Tests in the stripForgeDirs describe of test/crew.test.ts; without the boundary the test fails (tried once).
• #833's write-time re-check: the ship verb asks about a hold again at the write. The test plans a row (nothing to refuse), sets a hold, then writes: the row says who holds it and the pull is not merged. Without the re-check the test fails (tried once).
• docs: docs/ARCHITECTURE.md says commit, like version, is on the public health route, and why that is acceptable.
• #583: test/apps-idle.test.ts waits for production to settle and, when it is not running, fails with the environment's own error in the message, so the next flake names its cause. It stays out of the gate; the issue stays open.

Ran under a heavy slot on the merged head: tsc clean; crew, asks-bulk, apps-idle, holds, reserved-names, studio, shards, check-list and hollow in one call, 318 pass, 0 fail.

🤖 Generated with Claude Code