ci: write the GKE runner as the bare scale-set name - #571

The proof jobs of ci-post.yml, corpus-parity.yml and native-live.yml name their
runner pods with `runs-on:` plus a `group: gke-runners` / `labels: gke-...`
mapping. The org CI canon accepts a GKE runner only as the exact bare scalar
scale-set name (`runs-on: gke-8cpu-16gb-amd64`): no mapping, array,
`self-hosted` label or expression. All 11 jobs now use the scalar with the same
size and architecture as before; the runner group still restricts which
workflows may start on it.

scripts/ci_preflight.py pins the exact scalar for each moved job and refuses any
other shape on a GKE name as a wrong shape: a group/labels mapping (block and
inline), a one-label array, a `["self-hosted", ...]` array, an expression, a
different letter case, and a different size or architecture. The per-label cap
(8 per label), the pull-request reachability refusal and the credential refusals
are unchanged and now read the scalar. Every rewritten test keeps its original
property; the real workflow files are the positive control.

docs/ci.md sentences replaced, on the same lines, named by the commit that
introduced each:
- c85a40b: "post preflight/parity use", "profiles use fixed", "fixed
  architecture/size labels", "literal VM arrays or exact", "`HOSTED_JOB_RUNNERS`
  pins hosted jobs and exact GKE mappings"
- 3ea5a0d: "The amd64 post build shards", "The four amd64 post legs"
- 3a0d150: the `ci-post.yml` and `corpus-parity.yml` table rows
- 8154993: the `native-live.yml` table row

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>