feat: don't reply to commands queued while the bot was offline (per-command opt-in)

[sg]

What changed

• Detection: messages the radio queued while the bot was down are exactly those drained between start_auto_message_fetching() and the first NO_MORE_MSGS event. The bot now marks that window; CommandContext.from_queue flags each message. Failsafe: window force-closes after 60s if the end marker never arrives.
• Per-command flag: new process_queued column in command_config (NULL = script default) + PROCESS_QUEUED module attr (default False = ignore backlog). Editable via web Manage→Commands (queued checkbox) and the existing PATCH API; shown as queued-ok in !adm command list.
• Skipped commands are still ingested/stored — only the reply is suppressed; skips are logged and counted in the drain-complete log line.
• First DB migration hook (DB._migrate): adds the column to existing databases.
• Docs (Bot_Usage, example-bot-command) + 9 new tests (suite: 104 passing).

Why

On startup the bot rapid-fire replied to hours-old queued commands; !path answered 'direct (no path)' since queued messages carry no routing path.

Breaking changes

Behavior: backlog commands are now ignored by default (opt back in per command).

🤖 Generated with Claude Code