Add inno comply: GDPR deletion verification and data-residue scanning

[ringo380]

Implements epic #165 (GDPR & Compliance Verification) - the last unbuilt piece of the v5.0-v5.2 roadmap. Epics 8/9/11 (undelete, timeline, binlog correlation) already shipped.

inno comply is the inverse of inno undelete: instead of recovering data that lingers, it verifies a value has been purged from every InnoDB-retained location in a file and reports residue where it remains.

What's included

• --verify-deleted --where col=value - decode-and-compare over live, delete-marked, free-list, and undo DEL_MARK records. --thorough adds a raw byte sweep of slack space using the value's InnoDB-encoded form.
• --scan-residue --pattern <text|hex:..> - literal byte-pattern sweep across all page regions (record heap, free space, header/trailer), reporting page/type/offset/region with hex context.
• --encryption-audit - encrypted vs plaintext page report and key availability.
• inno audit --compliance --pattern - directory-wide residue scan across every tablespace, in parallel.
• Docs - comply CLI reference + a GDPR-verification walkthrough guide, wired into SUMMARY, the CLI overview, and the sitemap.

Design notes

• Reuses the existing scan surface - scan_free_list_records, scan_undo_for_deletes, decode_page_records, walk_lob/record walkers - so no binary-format parsing was rewritten. Net-new code is the residue/verify orchestration, the value encoder, and the CLI.
• Honest scope - the scan covers only the files passed in, not the OS page cache, replicas, or other backups, and reports byte/record-level residue rather than certifying legal compliance. The docs and --help state this plainly.
• No new dependency - patterns are literal (UTF-8 or hex:), so the core library stays regex-free and WASM-lean.

Testing

• 30 new tests (13 unit + 17 integration) against a real MySQL 9.0 fixture with a known value.
• Both scan paths mutation-checked: breaking value_matches flips exactly the "present" verify tests; breaking find_all flips exactly the scan tests. Present/absent pairs are the guards.
• cargo fmt --check, wasm target check, and mdbook build/test all pass. New code is clippy-clean.

Closes #175, #176, #177, #182, #183.

Note: a pre-existing completions_test stack overflow and some newer-clippy-lint errors in unrelated files (simulate.rs, schema.rs, recover.rs, fts.rs) reproduce on master and are untouched here.

[ringo380]

Deep review (auto-fix)

Applied 6 fixes to the working tree (uncommitted - review with git diff, then commit and push):

#	Fix	Reason	Location
1	hex: pattern with non-ASCII input panicked the whole process (&hex[i..i+2] sliced across a UTF-8 char boundary); now validated as ASCII hex digits and returns a clean Argument error	bug (crash)	compliance.rs#L57-L65
2	--verify-deleted --where col= (empty value) reported every NULL row as residue and flipped fully_purged to false; empty value is now rejected	bug	comply.rs#L100-L105
3	--thorough relabeled every raw byte-pass hit raw_slack, so a live-record match was reported as unreferenced slack; now tagged raw_<region> (e.g. raw_record_heap) from the real region	bug / comment	compliance.rs#L525-L531
4	audit --compliance --csv wrote the free-form error field without csv_escape; a comma in an error message corrupted the row (health path already escapes/skips)	bug	audit.rs#L758-L761
5	Em-dashes in user-visible CLI output, rustdoc comments, and the published mdBook guide replaced with -	CLAUDE.md	comply.rs#L165-L169
6	comply.md opening overclaimed "purged from every InnoDB-retained location" (the default pass only walks the clustered index, not secondary indexes); reworded to its real scope, and the raw_slack docs corrected to match fix #3	comment	comply.md#L1-L4

Verified locally: cargo fmt --check clean, cargo build --release OK, all compliance/comply/audit tests pass (the relabeled verify_thorough_adds_raw_pass included), the panic input hex: non-ASCII now exits 1 not 101, and the changed files are clippy-clean.

Could not auto-fix 2 issues (left for manual review)

• --encryption-audit reports key_available from encryption_info().is_some(), which is identical to is_encrypted() - so it says "key available: yes" even with no --keyring. The honest signal is decryption_ctx.is_some(), which has no public getter. Fix needs a small accessor on src/innodb/tablespace.rs (outside this PR's files) plus a decision on the intended semantics - skipped rather than guess.
• audit --compliance --csv on an empty data directory prints the plain-text "no files" message instead of a CSV header. This is the pre-existing shared empty-files block (same gap PR Complete remaining v3.2 issues: depth, prometheus, watch events, web audit #118 fixed once for --prometheus); it is not on lines this PR added, so it's out of scope for an auto-fix here.

_{- Fixes are uncommitted in the author's working tree. Review before merging.}