chore(rivet): close RQ-64-DEPS and record WHY RQ-64-ARCHMODEL defers

[avrabe]

v0.64 release prep — two artifact dispositions, no code

RQ-64-DEPS → implemented

Both done-when clauses discharged. The exit condition landed in #1186; all four
held bumps now carry a disposition on the PR itself:

PR	outcome	reason
#1106 wit-parser	merged	parsing layer; cannot reach selector or encoder
#1107 rivet action	merged	one line in a non-required workflow — the rivet binary is pinned separately and untouched
#1108 z3	closed	breaks required Bazel Build & Proofs; green on main and on 3 siblings sharing the base
#1110 wast	merged	Kani red was Install Kani — environmental, still blocked, cleared on re-run

Four bumps, four outcomes, each on its own evidence. That spread is the argument
for the exit condition being a procedure rather than a policy: "0.x-minor is
breaking, so hold" produces the queue; only per-bump evidence says which way each
one goes.

#1108 is the first time the enforced hold has stopped a breaking 0.x-minor
before it landed. The rule was written after ordeal 0.9→0.12 auto-merged as a
"minor" and hung CI for days (#849), and had never been falsified until now.

RQ-64-ARCHMODEL stays proposed, with its reason attached

Following the v0.63 precedent (35402926). spar#445 is still open, no
activity since 2026-09-03 — re-verified at cut time, not carried forward.

This is an external blocker, categorically different from a scope deferral:
nothing went stale, nothing got harder. It cannot proceed because the tool it
depends on silently accepts input it should refuse, and building an
architecture model on a parser that won't tell you when you got it wrong produces
exactly the unfalsifiable artifact this programme exists to avoid.

Fourth consecutive release recording feature-loop steps 1-2 as N/A (#1136). The
artifact notes something easy to miss: the conformance gate does not accept
that prose as evidence. It derives NA-FILED only from a release-scoped
artifact existing — so the N/A is discharged by filing, not by asserting "synth
is a Rust compiler, not AADL-architected". That assertion is true, and has never
been examined, which is the point of #1136.

Verified: rivet at main's 40-error baseline with 0 broken cross-refs,
status_evidence exit 0, claim_check 62/62.

Refs #965, #1136

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!