feat: stateless nREPL port auto-discovery via process scanning

[licht1stein]

Closes #20

Summary

When no .nrepl-port file, -p flag, or BREPL_PORT env var is found, brepl now automatically discovers running nREPL servers by scanning OS processes. Zero configuration, fully stateless.

How it works

1. Runs lsof to find TCP ports listened on by Java/Clojure/Babashka processes
2. Checks all candidate ports in parallel (pmap)
3. Each port is validated with a single TCP connection — sends describe to confirm nREPL, then eval to get the server's working directory
4. Returns the first port whose CWD matches yours

Port resolution order (updated)

1. -p <port> CLI argument
2. .nrepl-port file (walks up from file dir for -f)
3. BREPL_PORT environment variable
4. Process scanning ← new

Performance

• Real nREPL servers respond in <5ms on localhost
• Non-nREPL ports rejected within 100ms socket timeout
• All ports scanned in parallel — discovery completes in ~100-200ms even with many Java processes

Changes

• src/brepl/lib/discovery.clj — new module: get-listening-ports, parse-lsof-output, gather-port-info, discover-nrepl-port
• src/brepl.clj — discover-nrepl-port added as final fallback in resolve-port's or chain
• README.md — documents process auto-discovery in Port Configuration section

Design decisions

• Stateless: no caching, no session files, scans fresh every time
• CWD matching only: if the nREPL server's user.dir matches your CWD, it's a match
• No new CLI flags: discovery is automatic and transparent
• Graceful degradation: if lsof fails or no matching port found, falls through to error message

Inspired by clojure-mcp-light's discovery approach, with improvements: single-connection validation, parallel scanning via pmap, IPv6-aware lsof parsing, and Clojure-side command filtering (no shell grep).