Fix #65176: Auto-terminate live gateway on platform lock conflict

[AllardQuek]

When a new gateway detects a platform lock conflict (e.g. Telegram bot token already in use) and the holder is a live gateway process, it now terminates the holder and retries the lock acquisition instead of retrying indefinitely with a fatal error.

This mirrors the --replace behavior but applies it at platform-connect time when the scoped token lock detects a conflict, resolving the issue where the gateway would wait for manual SIGKILL of the conflicting process.

Changes:

• gateway/platforms/base.py: Modified _acquire_platform_lock() to detect live gateway holders via _looks_like_gateway_process(), write takeover marker, force-terminate with terminate_pid(), wait 0.5s, and retry lock acquisition. Falls back to fatal error only if retry still fails.
• tests/gateway/test_platform_lock_auto_terminate.py: New test file with three test cases covering termination of live gateway holders, non-termination of non-gateway holders, and fallback to fatal error when termination fails.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #65178: both implement adapter-level takeover of a live scoped platform-lock holder for #65176. #65178 is the earlier open implementation and includes explicit ownership/identity safeguards; this PR should be considered there instead.