fix(core): stop the host API from starting at login once core is down

[arthware-dev]

Summary

• stack down core left the dev.famstack.api LaunchAgent in ~/Library/LaunchAgents with RunAtLoad, so the next login started the host API on 42001 while core was reported down. After stack destroy core or stack uninstall it pointed at a deleted wrapper and launchd kept respawning it.
• core's stop hook now unloads the agent and removes the plist; stack up core writes it again. A plist with the same label that starts another instance's wrapper is left alone.
• The remote e2e runner no longer deletes the plist by hand after a run. The install lane's pristine reset still does, for crashed runs.

[arthware-dev]

Both problems confirmed on the remote test Mac before the fix, on the previous main, in a scratch instance:

• up core, then down core: the plist stayed, agent not loaded, 42001 closed. launchctl bootstrap gui/<uid> of the leftover plist (what a login does) loaded the agent and 42001 answered again.
• up core, then destroy core: the plist stayed, pointing at <data_dir>/core/famstack-api, which destroy had deleted. Bootstrapping it: state = spawn scheduled, last exit code = 78: EX_CONFIG, launchd retrying under KeepAlive.

Unit: tests/unit/stacklets/test_core_launchagent.py drives the real hooks through the framework's down and destroy with a temporary home and a recording launchctl stub. 4 of its 6 tests fail against the old stop hook. make lint and make test-unit pass locally.

[arthware-dev]

After the fix, on the remote test Mac, in a scratch instance:

Step	plist	agent loaded	42001
up core	present	yes	open
down core	absent	no	closed
down core again	absent	no	closed
up core	present	yes	open
uninstall --yes	absent	no	closed
uninstall --yes again	absent	no	closed

The first-run install lane on this branch (script/test remote install fix/core-launchagent, at cf591ea): 10 passed, 1 xfailed (the known "printed command on PATH" case), and the reset left the Mac pristine.