Generate command help and shell completions from shared command metadata

[Widthdom]

Summary

Eliminate drift between accepted flags/subcommands, usage grammar, next-step suggestions, and generated shell completions.

Current v1.38.0 evidence / reproduction

• --strict-not-found is accepted by excerpt/inspect but absent from command help and flag help.
• index --help omits accepted --parallelism and --no-progress flags.
• deps results recommend accepted --suppress-noise, --symbol, and --symbol-family flags that deps --help does not list.
• Search Usage makes recipe-only --include-query appear applicable to named queries.
• Zsh completion omits workspace clear and workspace deactivate, both present in help.

Impact

Users cannot discover supported controls and can construct invalid commands from authoritative-looking help.

Suggested direction / acceptance criteria

• Model per-command flags, exclusivity, usage forms, and subcommands in one schema.
• Generate help, usage, next-step flags, and shell completion from that schema.
• Add a consistency test comparing parser acceptance with rendered help/completion inventories.

Additional consolidated evidence

status --json exposes sqlite_connection_policy, but status --explain sqlite_connection_policy --json rejects that field even though it is an operator-facing readiness and risk contract.

Additional acceptance criterion

• Generate the status explain registry from the same metadata as emitted status fields, including sqlite_connection_policy.

---

Dogfooding candidate IDs: Q029, Q037, R003, R006, R008, M077. Verified against cdidx v1.38.0 at commit 2919bfc9b1b83cfa7ae1d964845d0f9ffca002d1.

[Widthdom]

Work started on branch fix-issue4571-4575-4621.