AI spam: classify captions, log alert delivery, add /testalert

[TayoBus]

Context

Porn spam (forwarded channel posts, payload in the media caption) bypassed the AI spam monitor: the filter and handler only inspected message.text, so caption-carried spam never reached classifier.dev. Separately, alert delivery failures were silent — a misconfigured ai_spam_alert_chat_id produced no visible signal.

Changes

Caption support (handlers/ai_spam_monitor.py)

• New _classifiable_text() helper returns message.text or message.caption or "".
• ai_spam_filter and handle_ai_spam_monitor now match/classify text or caption (still skips commands, short texts, bots, admins/trusted).

Alert delivery logging (handlers/ai_spam_monitor.py)

• Warning when no alert chat is configured for the group (previously a silent return).
• Info log with the target chat_id before sending, success confirmation after.
• Send failures now include the chat_id in the error log.

New /testalert DM command

• Admin-only, DM-only: /testalert sends a clearly-marked test message to the configured ai_spam_alert_chat_id of every group the caller admins (or one group via /testalert <group_id>), then reports ✅/❌ per group in the DM with the chat id used and the error detail on failure.
• New handlers/testalert.py + plugins/builtin/testalert.py, registered in manager.py/definitions.py (MANIFEST_ORDER); not guard_plugin-gated, same as /status.

Verification

• 1299 tests pass (pytest), including 9 new /testalert tests and 6 new caption/filter tests; test_plugin_manager expectations updated for the new plugin.
• ruff check and mypy clean.
• Classifier sanity-checked live against the reported spam text: explicit 0.98.

Note

Bot API chat_id for a group is the numeric group id, so ai_spam_alert_chat_id should be the (negative) admin group id with the bot as a member — /testalert confirms exactly this.