Retire the pr workflow Bats suite where Go covers it #923

Problem

tests/skills/pr/ holds the workflow Bats suite: 52 cases in about 699 lines that exercise the pr entry scripts end to end (preflight, review, publication, scaffold, validation) against the frozen shell. Once child 12 forwards those entry points to Go and children 01 to 09 carry Go tests for the same behavior, most of these cases test the shell twice and slow the Bats jobs on Ubuntu and macOS. Each case needs a recorded decision: covered by a named Go test and removed, or kept because it proves something only the real scripts can show.

Part of #588 and #323. Changes only tests/skills/pr/ workflow suite files and the opt-in flag RUN_HEADLESS_REVIEWS. No shell body, library or generated file changes.

Acceptance criteria

1. AC1: Given the 52 cases before the change, when the change lands, then a table in the pull request lists every case with its disposition: removed with the name of the Go test that covers it and the name of the retained frozen-path case that covers the same shell outcome, or kept with the reason (including "frozen shell outcome with no other coverage: kept until the shell is removed").
2. AC2: Given each of the five forwarded entry scripts, when the suite runs, then at least one frozen-path case per script remains, running the script with a binary that lacks the behavior, and no case is removed whose only coverage of a supported frozen-shell outcome (a refusal, an exit status, a message, a file written) is itself.
3. AC3: Given the pre-push hook and a real git push to a bare remote, when the suite runs, then the cases that exercise them are kept.
4. AC4: Given the forwarding cases for exec, argument pass-through and exit-status forwarding, when the suite runs, then each remains at least once for each forwarded entry script.
5. AC5: Given RUN_HEADLESS_REVIEWS, when the suite runs with it unset, then the delegation case that uses the deterministic reviewer still runs as today (it is not skipped), and when it is set, then the retained live-review cases also run; the flag is documented where it is read.
6. AC6: Given the Bats jobs on Ubuntu and macOS, when they run on the merged head, then they pass and the suite's case count is lower than 52 by the number of removed cases.
7. AC7: Given every removed case, when its behavior is read in the Go tests it names, then the Go test asserts the same outcome (exit status and message), checked by a reviewer for each row of the table.
8. AC8: Non-functional: no shared helper under tests/skills/pr/ is changed by this pull request.

Scope and non-goals

Scope: the workflow suite under tests/skills/pr/ and the table of dispositions.

Non-goal: the compatibility, section-pipes and writer suites, which child 15 retires. Non-goal: the agent-pr suites and the shared helper, which child 16 handles. Non-goal: changing or deleting any shell. This pull request deletes only Bats cases: no script, library or generated copy is removed, and no owner-tier path in .agents/agent-skills.conf is touched (those go to the single cutover issue under #588). Non-goal: removing cases whose behavior has no Go test, or whose frozen-shell outcome has no other retained coverage; those stay and are listed.

Approach

Meet AC1 to AC8 by building the disposition table first, with each Go test name resolved to an existing test from children 01 to 09, then removing the covered cases in one pull request. Cases with no Go counterpart stay and are listed as follow-ups.

Dependencies

Blocked by: #921.

Prerequisite: children 01 to 09 merged (reached through child 12).

Size

M. One pull request, mostly deletions: about 699 lines of Bats at most are touched, plus the table in the pull request body; under the code line limit.

Verify plan

• AC1, AC2, AC7: the disposition table in the pull request, checked row by row at review, including the frozen-path coverage column.
• AC2, AC3, AC4: the remaining case list shows the named cases, and the Bats job passes on both operating systems.
• AC5: the suite run with the flag unset (asserting the deterministic-reviewer delegation case ran and passed) and set (asserting the live-review cases ran).
• AC6: the Bats jobs on Ubuntu and macOS on the merged head, with the case counts before and after.
• AC8: the diff path listing shows no helper file.
• Definition of done: D1 each AC is verified at the merged head, with the commands and results cited in the pull request's Verification. D2 an independent review approves the exact merged head, and every finding has a recorded disposition. D3 required checks pass with no gate loosened. D4 the leak scans the repository specifies are clean, and every post made for this work passes the per-post leak review before it is posted. D5 every commit is signed. D6 the documentation matches the change or the Verification says why there is no impact. D7 every commit has an Assisted-by trailer and the pull request has the provenance:agent-assisted label. D8 the pull request stays within the size limits. D9 follow-ups are filed as issues and listed under Related. D10 the merge closes the issue through a closing reference, the board shows Done, and the final Status: comment names the merge commit.

Risks

A removed case may have covered a behavior the Go tests miss, which the row-by-row review in AC7 is meant to catch. Undo: revert the change; the cases return and run against the forwarders.

Operator authority

None.