Decide: normalise the ~42 non-prettier-clean tracked files, and whether format:check becomes a CI gate

[steilerDev]

[orchestrator] Split out of #2024 on the product-owner's scoping advice: #2024 bundled a two-line safety fix with two genuinely larger debates. #2024 keeps the safety fix (add wiki/ to .prettierignore); this issue carries the debates so neither blocks the other.

Context

Roughly 42 tracked source files in this repo are not clean under the current Prettier config — union-type wrapping, array wrapping, a stray blank line in package.json, and similar. They sit on beta unnoticed because format:check is not run by any job in .github/workflows/ci.yml.

The consequence is felt by anyone running repo-wide npm run format, which every implementing agent does before handback per the Local Validation Policy: the command reformats ~42 files that have nothing to do with the change in hand. During one batch session this had to be untangled on every single item, by classifying each modified file as drift-or-real before committing.

The two decisions

• 1 Should the ~42 files be normalised in one dedicated, behaviour-free commit? Leaving them means every future repo-wide format run produces spurious diff noise that a human or agent has to classify by hand — and the classification is easy to get wrong in the direction of committing unrelated churn. Normalising them means one large but reviewable commit that touches no behaviour. State the decision and the reasoning either way.
• 2 Should format:check become a CI gate? If yes, note it can only be meaningful after decision 1, since it would fail immediately today. If no, record why, so the next person does not re-propose it — the absence of a gate is currently indistinguishable from an oversight, which is how it stayed latent.

Notes

• Reliable technique for separating real changes from formatting drift, for whoever works on this: for each modified file, compare prettier(git show HEAD:<file>) against the working copy; identical means the file's only change is drift and it can be reverted wholesale. This is how the ~42 were identified.
• Interacts with npm run format reformats wiki/*.md, dirtying the submodule on every repo-wide run #2024: if wiki/ is ignored there, the wiki markdown is out of scope for both decisions here.