fix: release hdc fport so the local port is free after each run

[tonycoder-hub]

Fixes codematrixer/hmdriver2#58: after each use the hdc fport mapping was left in place, so the local port stayed occupied.

Cause

hdc fport tcp:<lport> tcp:8012 is opened when HmClient.local_port is first used. The only teardown path was Driver.__del__ → HmClient.release() → hdc fport rm. __del__ is not reliable at process exit, and _rm_local_port() read the local_port property (which opens a new forward if none existed). There was no public close() / context manager / atexit cleanup.

Change

• Track the allocated local port explicitly; teardown never opens a new fport.
• Driver.close() / stop() / __enter__/__exit__ / atexit / __del__ all call the same idempotent release that runs hdc fport rm.
• hdc fport rm uses a 5s timeout so a hung fport rm cannot block process exit.
• Closing one serial no longer clears every Driver singleton (_instance.clear()).

Related: codematrixer#47

codematrixer/hmdriver2#47 (scripts hang after finish) is partly the same lifecycle. Missing teardown left the forward around (codematrixer#58). When __del__ did run, hdc fport rm could block forever inside _execute_command (also codematrixer#54). The timeout on fport rm addresses that hang; codematrixer#47 may still need follow-up if something other than fport rm is holding the interpreter.

No Harmony device required. Added mocked tests in tests/test_hdc_port_lifecycle.py.

This PR is on the fork only (not opened against upstream). Does not redo codematrixer#59 lxml, codematrixer#60 current_app, codematrixer#61 textContains, codematrixer#62 foldable dump.