fix: kill stale MCP processes on reconnect to same endpoint

[yuvrajangadsingh]

Problem

When MCP clients reconnect (e.g. running /mcp in Claude Code), each reconnect spawns a new chrome-devtools-mcp process without killing the previous one. Multiple CDP clients on the same debug port cause Network.enable timed out errors because sessions conflict.

Reproducing is easy: run /mcp three times in a row and check lsof -i :9222. You'll see 3+ node processes all connected to the same debug port.

Fix

Adds endpoint-based PID lock files. On startup, the server:

1. Checks if another instance is already connected to the same endpoint
2. Sends SIGTERM (with SIGKILL fallback after 1s)
3. Waits for the old process to actually exit
4. Acquires the lock with its own PID (atomic via openSync with wx flag)
5. Releases the lock on exit (SIGINT/SIGTERM/exit handlers call process.exit())

Lock files are keyed by normalized endpoint URL (not just port) so different hosts on the same port don't collide. Both --browserUrl and --wsEndpoint connections are covered.

Changes

• src/browser.ts: acquireEndpointLock() and releaseEndpointLock() functions
• src/bin/chrome-devtools-mcp-main.ts: early lock acquisition on startup, cleanup on exit

Tested

Verified locally on macOS with Arc browser (port 9222):

• Instance 1 acquires lock, Instance 2 kills Instance 1 and takes over
• Lock file correctly transitions between PIDs
• Old process actually dies (SIGTERM + SIGKILL fallback)
• Lock released on clean exit

Fixes #1763
Related: #1156, #657