configure.sh model inference on carries; harden ruleset snippets - #334

Fixes the operational-carry gap and the snippet nits bundled into #332
(raised by Copilot on the Vantage-Config carry and the #330 sweep PRs).

1. configure.sh: when the registry is absent (a downstream carry), the
   model is inferred from which develop payload is carried - a carry
   holds exactly its own model's. Ambiguous layouts (both or neither,
   e.g. a partial copy) abort rather than guess. The documented no-arg
   form now works on operational carries. The model may also be passed
   as the sole argument (configure.sh operational); previously arg 1
   was always parsed as a repo name.

2. repo-config/README.md regen snippet + AUDIT.md section 6 ruleset
   snippet: enforce exactly one ruleset per name before using the id -
   a duplicate name previously produced a multiline id that silently
   corrupted the follow-up API path; zero/duplicates is declared drift,
   so fail loudly (regen) or report and continue (audit). Single list
   fetch, jq-selected.

3. repo-config/README.md documents the inference fallback and the
   model-only argument form.

Declined from the bundle with rationale: gh api --input - portability
(fleet standard is current gh; no temp-file churn) - recorded on #332.
The jq-derived secret lists and --paginate hardening apply to the
blessed downstream self-audit reference, which lives in the
Vantage-Config carry - propagated there, not here (the hub's fleet
audit reads spec/secrets.json programmatically already).

Resolution logic matrix-tested (infer release/operational, ambiguous
both/neither abort exit 1, explicit arg override, model-as-sole-arg,
unknown-model abort); hardened snippets live-tested read-only against
this repo (both rulesets in sync; duplicate simulation counts 2).
shellcheck and bash -n clean; markdownlint and editorconfig-checker
clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

gh api --paginate emits each page as a separate JSON doc; re-assemble
with jq -s so a name match on a later page is never missed - the same
trap configure.sh already guards against. Live-tested read-only against
this repo (both rulesets in sync).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…lot #334)

With --jq '.[]' the paginated fetch emits one JSON object per ruleset
across all pages; jq -s re-assembles them into the single array the
selections expect. The old wording described the un-filtered per-page
behavior.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…334)

An operational repo's develop ruleset diffs against
operational/develop.json - the same registry-workflowModel selection
audit.py makes; the snippet hard-coded repo-config/$b.json and would
false-DRIFT every operational repo. Live-tested read-only against
Vantage-Config (operational: both rulesets in sync via the swapped
payload).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The model-only invocation examples now use the repo-root-relative
repo-config/configure.sh form the docs use elsewhere, and the payload
inference tests use -f (regular file), not -e. Inference matrix re-run:
release/operational infer, both/neither abort. shellcheck clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…#334)

The round-4 patch rewrote repo-config/README.md through a text-mode
round-trip that stripped its CRLF endings - the Lint sources job
correctly failed on it; restored byte-exact CRLF (editorconfig-checker
clean). The regen snippet's develop-payload presence check now uses -f
like the configure.sh inference checks.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>