docs: track CLAUDE.md instructions in the repo

[juemerson-at-purestorage]

Summary

CLAUDE.md (repo root), .github/workflows/CLAUDE.md, and Tests/CLAUDE.md
existed only in local working trees -- never committed on any branch, despite
the root file describing itself as repo-committed guidance safe to publish.
This adds all three to version control so they're actually visible to
collaborators.

Also adds one new section to the root file: regenerate Reports/ drift-report
artifacts in the same commit whenever cmdlet parameter/endpoint coverage
changes, and merge/rebase onto current main first (to avoid mixing in an
unrelated spec-version bump) -- a real gap found finishing PR #51 this session.

No code changes.

Test plan

• Docs-only change, no functional/test impact

[juemerson-at-purestorage]

Rebased onto current main (93f0ffd) and updated. The original commit was authored 116 commits back, before the issue-#63 CI work landed, so the files described a repo whose test entry point and gates no longer matched. Nothing in them was wrong — they were silent about the parts most likely to red a build.

Added, in a second commit so the rebase and the new content stay separable:

Tests/CLAUDE.md

• The CI entry point, with the correction that -Edition selects which coverage baseline to measure against rather than which interpreter runs — so each edition needs its own host, and -Edition winps51 from pwsh 7 measures a 7 run against the 5.1 ceiling.
• The coverage gate: MaxSkipped as a ceiling with headroom rather than a pin, and RequiredDescribes for blocks that skip gracefully and so contribute neither a pass nor a skip — which a skip ceiling cannot see disappear.
• Why the two editions differ by ~190 skips by design, the gitignored tools/specs cache the tooling tests depend on, and -Confirm:$false for ConfirmImpact = 'High' cmdlets under -NonInteractive.

.github/workflows/CLAUDE.md

• pull_request runs test the merge commit: a green run goes stale when main moves with no branch change, and a manual re-run replays the recorded SHA rather than re-measuring, so it cannot clear a stale result. Both were hit during Fusion context_names Phase 1: session-scoped context and client-side validation (#25) #109.

CLAUDE.md

• A before-you-push section pointing at both, and the standing rule that version bumps and CHANGELOG.md edits are a separate maintainer decision rather than part of a feature PR.

Still additive only — no existing file is modified, since none of these three exist on main.