Feature: caseworkctl audit verify — offline verification of the hash-chained audit journal

### Context

The audit journal is an append-only hash chain of envelopes (`envelope_id`, `timestamp_unix_ms`, `prev_hash`, `record`, `record_hash`; see `crates/registry-platform-audit`). Operators need to verify a journal offline in two situations:

- after restoring an instance from an export, to confirm the chain is intact and continues from the exported head;
- during incident review, to demonstrate the journal was not truncated or rewritten.

Nothing shipped does this today — verification requires re-implementing the chain derivation from the crate source. I had to write an independent verifier to prove chain continuity across an export/restore cycle (it reproduced the runtime's hashes and confirmed post-restore appends extended the exported head), which is exactly the check operators should get out of the box.

### Ask

Add `caseworkctl audit verify` that:

- takes a journal path plus the audit master key (via a `secret:` reference or env),
- re-derives and checks every record hash and prev-hash link,
- prints the verified record count and head hash, and exits nonzero on any mismatch,
- optionally accepts a known head (`--from-head `) to verify continuation from that head rather than from genesis — the export/restore case.

Implementation can reuse `registry-platform-audit`'s hashing directly rather than a parallel implementation.

Start with the caseworkctl audit command entry point and the hashing implementation in crates/registry-platform-audit, then trace how journal paths and secret: or environment references are handled. Define completion as verifying every record and prev-hash link, supporting --from-head, printing the record count and head hash, and returning nonzero on mismatch.

Written by the indexing model from the issue text.