Last state of the private idalib worker backend

The backend ida-tui used before the IDA Code Mode port: idatui/worker.py +
worker_client.py (a private subprocess speaking length-prefixed pickle over a
unix socket) with server/patch_server.py injecting our extra tools into
ida-pro-mcp.

Kept as the A/B reference. The port replaces the transport wholesale, so this
is the tag to check out when asking 'is this a regression or has it always
done that?' -- which is exactly how the xref ordering, the trace-follow race
and several performance claims were settled. tests/run.py works here, and
/tmp/cmport/bench.py (if still around) picks whichever client the checked-out
tree has, so it benchmarks this tag unmodified.

Reference numbers on targets/echo, this box: heads(200 rows) 2.65ms,
decompile warm 30.0ms, decomp_map 45.8ms, full pilot suite ~60s.