fix(scripts): the gates die on the bash macOS actually ships

[ruslan-korneev]

Closes RUS-188.

The problem

macOS ships bash 3.2.57 as /bin/bash, and #!/usr/bin/env bash finds a newer one only when
Homebrew's is on PATH. Three core scripts used bash 4+ syntax:

Where	Construct	What actually happens on 3.2
gate.sh	${have,,} / ${want,,}	bad substitution — fatal in a non-interactive shell. gate.sh workspace with INTAKE_WRITEBACK=true aborts. Not a failing gate: no gate at all.
intake.sh	${kind^^}	same, in the writeback path
review.sh	mapfile	not a command → array unset → next line reads ${engines[$idx]} under set -u and aborts. A review that dies before spawning a reviewer.

bash -n is clean on all three, which is why this never showed up until something ran.

The fix

lib.sh gains lower/upper (tr-based) — case folding is needed in two scripts now — and review.sh
reads the engine lists with while IFS= read -r.

This closes the last known gap in the suite: check_xfail 'core scripts avoid bash-4-only constructs'
is promoted to check. A new assertion exercises lower, upper and the array read under
/bin/bash itself rather than trusting the replacements.

The detector was broken too

It shared a comment-and-quote stripper with the GNU-tool rule. But "${have,,}" sits inside double
quotes nearly every time it appears, so stripping quoted strings blinded the check to exactly what it
was looking for. Split in two:

• comments only → bash-4 expansions
• comments and quoted strings → tool names, since naming a binary in a warning is not a use of it

How to verify

bash tests/run.sh                            # 145 passed, 0 failed, 0 known gaps
ADW_TEST_SHELL=/bin/bash bash tests/run.sh
/bin/bash --version | head -1                # 3.2.57 on macOS

Mutation, one construct at a time — each fails exactly one assertion:

python3 - <<'PY'
s=open('scripts/gate.sh').read()
open('scripts/gate.sh','w').write(s.replace('"$(lower "$have")" != "$(lower "$want")"','"${have,,}" != "${want,,}"',1))
PY
bash tests/run.sh portability     # 1 FAIL
git checkout scripts/gate.sh

Same for ${kind^^} in intake.sh, mapfile in review.sh, and breaking lower().

Worth knowing: the first attempt reported zero failures for two of them — the perl one-liners
doing the mutation had silently not applied, so that run proved nothing. The mutation step now fails
loudly when its own edit does not match, and that is the only reason the quote-stripping blind spot
above was found rather than shipped.

Not verified

• The two runtime paths that were actually dying — gate.sh workspace with INTAKE_WRITEBACK=true, and
  review.sh past engine selection — still are not covered end to end: one needs a tracker, the other
  needs a live engine CLI. What is covered is that the constructs are gone and that their replacements
  behave correctly under 3.2.
• Pre-existing and not touched here: if engines.sh pick-review ever returns nothing,
  ${engines[$idx]} is still an unbound-variable abort under set -u. Independent of the mapfile
  change — it read the same way before. Not filed yet; say the word and I will.