fix(readiness): exclude the resurrecting PyAutoCTI from the release gate

[Jammy2211]

Problem

Heart's main was red — 27 test_readiness.py failures + the polled-count test. Bisect points at 358ef2f ("config: poll the resurrected CTI repos"), not any readiness code change:

358ef2f^ : 2 passed
358ef2f  : 2 failed

That commit added PyAutoCTI to the config libraries group so the tick would poll it. But heart/readiness.py derives the release-validation gate set (_GATE_SHA_LIBS) and the library-scoring default from that same group. So every verdict began demanding a matching commit_sha for PyAutoCTI — and CTI is mid-resurrection (not released), so it has no release evidence to confirm. Result: a permanent release validation partially unconfirmed (unknown HEAD: PyAutoCTI) stale (−12 on every score), and the 5-classic-lib fixtures never matched → 27 failures.

Fix — separate "polled for health" from "gates the release"

• config/repos.yaml: PyAutoCTI carries release_gate: false (still polled by the tick; not part of the release gate). Flip to true / drop the key when CTI ships.
• heart/readiness.py: new load_release_gate_names() = polled libraries minus release_gate: false. _GATE_SHA_LIBS and the libs default now use it; load_library_names() stays the full polled set. Absent key ⇒ gating (default) — only an explicit false opts out.
• tests: polled-count sanity 22 → 25 (CTI added 3 repos across groups); new test_cti_polled_but_not_release_gating.

Verification

python3 -m pytest → 287 passed (was 27 failed / 259 passed on main).

Notes

• This failure pre-dated the docs PR docs: unify the call chain on Build (#84) #90 (a one-word AGENTS.md edit) and is unrelated to it — surfaced while merging the organism docs sweep.
• The design call (a resurrecting library is polled but not release-gating) is worth a maintainer eye — flagged here rather than buried.

🤖 Generated with Claude Code

---

Generated by Claude Code