feat: resilient violation reporting — DB spool + cron retry (#7)

[cport1]

Upgrades the minimal shutdown-flush reporter (shipped with #5) into the resilient delivery pipeline #7 asks for — without adding page latency.

What changed

• Spool table — new webdecoy_violation_queue (activator + DB_VERSION → 2.2.0, dropped on uninstall). Violations are written the moment they occur.
• Immediate, latency-free delivery — on request shutdown, fastcgi_finish_request() hands the response back to the visitor first, then a blocking batch POST drains the spool. So enforcement stays near-instant (a tripwire hit reaches the deny-list in seconds) while the visitor waits for nothing.
• Survives outages — a failed batch is retried once, then dropped; the queue is hard-capped at 1000 rows (oldest discarded) so a prolonged ingest outage can't grow the table unbounded. A webdecoy_flush_violations cron (every 15 min) drains whatever the shutdown path couldn't.
• Clearance preserved — the wd_clearance token survives the spool round-trip, so the tripwire → durable-deny loop still works even after a retry.
• Cloud-gated — no API key means no spooling and no reporting; rules still enforce locally.

Why a DB spool (not just fire-and-forget)

The P0 reporter used a non-blocking POST — fast, but it couldn't tell whether delivery succeeded, so a momentary ingest blip silently dropped detections. Spooling + a confirmed (blocking, post-response) send + cron retry makes delivery reliable without ever blocking the visitor.

Tests

Verified with a stubbed $wpdb + HTTP harness: enqueue, success clears the queue, clearance preserved through the spool, failure keeps the row and bumps attempts, drop after the retry, delivery on recovery before the drop, and the hard cap trimming overflow. Existing SDK suite still green (35 assertions).

Closes #7. Part of #16.