[callumreid]

• Add six python-metric-drafts operations for immutable draft creation, retrieval, historical previews, cancellation and publication with an approved fingerprint. Include source-file input, structured output and agent next actions.
• Add the persona --interruption-instructions-enabled flag while preserving explicit false and omitted PATCH fields.
• Reconcile the published API catalog: 125 of 192 operations have first-class commands, with 67 existing reviewed exceptions. All 464 top-level request fields on covered operations are modeled. The audit now rejects unparsed JSON-body calls; previously the three new draft request bodies silently escaped field checking.
• Keep this implementation PR at version 0.9.2. The 0.10.0 release bump will be a separate manifest-only PR after this merges; user documentation follows the release.

Validation: regression tests reproduced the three skipped request-body operations and the silent unsupported-call fallback at d834cdb before the fix. Afterward, 65 automation tests, cargo fmt, cargo clippy, Python lint/format, release-manifest validation and the live 464/464-field audit pass. The final Rust run passed 220/221 tests; the existing advisory update-notification test exceeded its unchanged 500ms wait in both the full run and isolated rerun while local host load was about 105. The initial Rust run passed all 221 tests before the audit correction. Current-head CI remains required. HTTP tests cover request bodies, local source-file errors, path-segment encoding and persona true/false/omission. No live draft creation, preview execution or publication was performed during validation.

Final-head CI at 97eed40f52a4956b276fb5cae8c0671eac9f1f6b passed all 221 Rust tests (including the locally timing-sensitive notification test), 65 automation tests, all five platform builds and CodeQL: https://github.com/coval-ai/cli/actions/runs/40357537345. Independent approving review remains the merge gate.