test(cli): the title-resume-hint gate could not see naive quoting

[Kyzcreig]

tests/cli/test_exit_summary_resume_hint.py gates cli.py:17984, and its
only fixture title was "My Cool Session" — no apostrophe, so
shlex.quote(t) and a naive hand-rolled f"'{t}'" emit BYTE-IDENTICAL
output for it. The test therefore could not tell correct quoting from
naive quoting, and mutant M9 (naive single quotes at that site) survived
the whole file. M9 is a real defect: real bash REFUSES
hermes -c 'don't' with "unexpected EOF while looking for matching ''". The class is already gated at the twin site (main.py:2003) by tests/hermes_cli/test_cli_hint.py's quo'te` parametrization — this was a
missing discriminator, not a missing gate.

Replaced the literal-spelling assertion with a ROUND-TRIP oracle
mirroring the twin's: drive the real print site, hand the printed line to
a real /bin/bash (NUL-delimited), and require the argv back to equal
["hermes", "-c", <title>, "-p", "dev"]. Parametrized over 12 titles
including the apostrophe discriminator, expansion/substitution, control
operators, and globbing (with a hostile_cwd so the globs can actually
hit). A round trip cannot go stale when the site's quoting style changes
— which is exactly what caused t_0c5ac29a.

VERIFIED (worktree at fda1503; .venv/bin/python -P 3.11.15,
PYTHONPATH=, HERMES_HOME=/tmp/te484-home, -p no:randomly;
every mutant ast.parse'd VALID-PYTHON, every restore cmp'd
byte-identical):
M0 control -> 16 passed GREEN
M9 naive single -> 1 failed, 15 passed KILLED [quo'te]
M8a pre-#889 double -> 4 failed, 12 passed KILLED
M8b bare/unwired -> 12 failed, 4 passed KILLED
cli.py untouched (git status: test file only). ruff 0.15.10 clean.
tests/cli + tests/hermes_cli/test_cli_hint.py: 1 failed, 1561 passed —
the failure is INHERITED (test_resume_quiet_stderr, 1 failed/1433 passed
on the same worktree stashed clean; passes 4/4 in isolation).

Stacked on #889 (daedalus-opus/t_c9e1a012-hint-shell): hint_value at
cli.py:17984 exists only on that branch.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[Kyzcreig]

apollo/merge-pass 2026-09-23: rebased onto current main (clean, own commits only, no content change) so CI runs against the post-#915 main; landing via fleet-merge on the FR-pause lane once green.

[Kyzcreig]

🤖 merged-by: apollo · lane: fr-pause-0922 · gate: BYPASS: FR PAUSED by Ace ruling 2026-09-22 (state/fleetreview-pause-20260922.md); t_e484ec2d approved · why: argus approved; 16 hint tests pass; 37 checks green, clean; CI green; Apollo merge pass 2026-09-22