Ask the code gate which role came back unread, not what ratio it read at

[leaf-agent]

Problem

ci on main has been red on tests/test_render_gate.py::test_the_render_gate_reports_code_the_reader_cannot_tell_from_its_block since a8b6636 ("Improve code and diff presentation"), and it is still red on today's main — reproduced locally at 2d46d4a:

AssertionError: ['[light] code marked cm reads at 3.4:1 against the block it is set on']
assert []

The test pinned the exact contrast ratio the gate's finding prints. That ratio is a reading of the theme's own surfaces, so when the commit moved --pre-bg from #f5f2e9 to #f5f5f3 as part of the near-neutral code palette, the faint fixture's #8b8577 went from 3.3:1 to 3.4:1 and the startswith(... "reads at 3.3:1") filter came back empty. Nothing about the gate or the palette is wrong — the theme change was the point of the commit, and the same commit deleted the theme comment that was the 3.3 digit's source.

Solution

Both ratio-pinned assertions now ask which role came back unread rather than what number it read at. unreadSyntax emits that sentence only where read < 4.5, so the finding's presence is the claim the test is making; the digits proved nothing extra and made every palette change a failure of this test rather than of the page. The second assertion carried the identical brittleness on --syn-string and is fixed in the same pass rather than left to redden on the next palette move. The two fixture comments that restated the same theme-owned digits are updated to state the colour and not the reading.

Nothing is loosened that the cases were arranged to distinguish. The faint case still asserts that only the role the style touched is named. The tinted case still discriminates the per-surface reading from a per-role one, because that role clears the threshold on the block — a gate that stopped at its clean line reports nothing at all.

Testing

• The single test fails on origin/main (2d46d4a) and passes with this change.
• uv run pytest tests/test_render_gate.py — 69 passed.
• Mutation check on the assertion this change loosened: keying unreadSyntax's seen set on the role alone instead of role-plus-surface — the "one block colour taken once per role" bug the tinted case exists to catch — still fails the test, with the finding list empty. The gate was restored afterwards; this PR touches no runtime file.
• ruff format --check and ruff check clean on both files.

The other 29 tests red in that run predate this commit, and the two that appeared alongside this one (test_a_reader_arrives_at_what_they_left_rather_than_watching_it_arrive, test_a_label_press_keeps_the_controls_keyboard_standing) did not recur on the next commit's run and pass locally — races rather than regressions, so they are out of scope here.

How the failure was isolated

Failure lists from three consecutive ci runs on main, diffed:

run	commit	failures
33340551132	f8dffeb (parent)	30
33340636265	a8b6636	32
33340823726	650db7e (child)	31

Three failures are new at a8b6636; only this one recurs on the child run.

Arithmetic behind the move: #8b8577 has relative luminance 0.2360. Against #f5f2e9 (0.8880) that is (0.938)/(0.286) = 3.28 → 3.3; against #f5f5f3 (0.9119) it is 3.36 → 3.4.

grep for other pinned ratios across tests/*.py found only test_interact_document.py:3590, which computes its threshold rather than pinning a reading.

This follows the shape of #66, which replaced a pinned number with the relation it stood for.

---

Automated fix for failed run