Adds support for running each Feishu bot in its own Node.js process instead of
multiplexing all bots in one. Useful when:
- Restarting a single bot shouldn't disrupt others
- Bots have very different memory/CPU profiles
- Different bots need to be on different code branches/versions

Two coordinated changes:

1. src/config.ts — When BOT_NAME env var is set, loadAppConfig() filters all
   four platform arrays (feishu/telegram/web/wechat) down to just the matching
   bot. Throws if no bot matches. Existing single-process mode (no BOT_NAME)
   is unaffected.

2. ecosystem.config.cjs — Replaces the single 'metabot' entry with one PM2
   app per bot read from bots.json. Each app gets:
   - BOT_NAME=<name> for filtering
   - Distinct API_PORT (base + index*3) and MEMORY_PORT (base + index*3 + 1)
   - METABOT_DATA_DIR=~/.metabot/<name>/ for data isolation
   - MEMORY_DATABASE_DIR + META_MEMORY_URL pointing at the bot's own port
   - Per-bot log files (logs/<name>-{out,error}.log)
   - API_PORT_BASE configurable via .env (default 10001)

Single-process mode still works — just leave BOT_NAME unset and run with the
old ecosystem entry pattern.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds handling for an external workflow (e.g. an `mb switch <bot>` script or an
`/api/bots/:name/switch` endpoint) that wants to restart a bot pointed at a new
working directory while resuming a specific Claude session and surfacing the
last few messages back to the user.

Mechanism:

1. The external tool writes ~/.metabot/<bot>/pending-switch.json before
   restarting the bot:
     { chatId?, workDir, sessionId, recentHistory: [{role, content}, ...] }

2. On startup, src/index.ts reads the file and either:
   - injects sessionId immediately if chatId is known + pushes a notice card
     listing the recent history, OR
   - if chatId is unknown, defers the notice into bridge.pendingSwitchNotice
     to be consumed by the next incoming message.

3. handleMessage() consumes pendingSwitchNotice on the first message after a
   switch, injecting the sessionId before the message goes through the normal
   command/query pipeline.

4. The file is unlinked after processing so the next restart is clean.

This is the missing half of a project-switch feature: ecosystem.config.cjs
already supports per-bot data dirs, but until now Claude wouldn't resume the
prior session after a workdir change, and the user got no indication that a
switch had happened.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>