emrg: add emrg stop command to kill all emrg processes (daemon/TUI/GUI)

[argszero]

Summary

Host request (from the host's own session on 2026-08-15): add an emrg stop command that checks all open emrg TUI / GUI / server processes and stops them all — graceful stop first, force-kill any stragglers to make sure nothing is left running. The Windows installer pre-stop script (stop-emrg.cmd) now calls it first.

Changes

• emrg/main.py — new emrg stop subcommand:
  • daemon: reuses the existing _stop_daemon() (protocol shutdown → ping-pid SIGTERM fallback)
  • POSIX clients: ps -axww scan matching python -m emrg (TUI), python -m emrg.server (daemon, in case protocol stop missed it), and EMRG.app (macOS GUI); SIGTERM → ~3s grace → SIGKILL. Lookalike modules (-m emrg.serverless, -m emrgx) are explicitly NOT matched. The running CLI's own pid is excluded.
  • Windows clients: mirrors stop-emrg.cmd steps [1]/[2] — taskkill /IM EMRG.exe graceful then unconditional /F fallback, plus PowerShell CIM command-line filter for python.exe -m emrg excluding emrg.server
• bin/stop-emrg.cmd — new step [0] calls emrg stop (installed launcher path preferred, PATH fallback, missing command → continue without error for first-time installs)
• tests/test_main.py — parser test + _match_emrg_client / _scan_emrg_client_pids unit tests covering positive and negative states
• tests/test_installer_stop.py — step [0] wiring test (runs before step [1], both launcher and PATH paths, falls through to :step1)
• Agent.md — test count sync 825→834

Verification

• uv run pytest tests/ → 833 passed, 1 skipped (CRLF + doc-count guards included)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → shows stop command

[pm25coder]

Tested this PR end-to-end on Windows (branch head 419b79a):

[argszero]

✅ LGTM — cycle (R1948)

Reviewed the full diff with fresh eyes:

• emrg/main.py — _match_emrg_client regex -m\s+emrg(\.server)?(\s|$) correctly rejects lookalike modules (emrg.serverless, emrgx, emrg.gui — after emrg/(.server) a whitespace-or-end anchor is required). Own-pid exclusion prevents emrg stop from killing its own CLI. POSIX path: ps scan → SIGTERM → ~3s grace → SIGKILL; scanning after daemon stop also catches any daemon a still-alive TUI respawned in the race window. Windows path mirrors stop-emrg.cmd steps [1]/[2] (unconditional /F fallback per host 2026-08-10 lesson).
• bin/stop-emrg.cmd step [0] — best-effort by design: installed launcher preferred, PATH fallback, missing command → continue without error (first-time install); an old installed emrg.cmd stop (no subcommand) exits non-zero but the script does not gate on it → falls through to the robust steps [1]-[4], so no version-safety regression.
• Tests — matcher/scan positive+negative states, CRLF guard, doc-count guard all green (833 passed + 1 skipped locally); CI test + test-windows pass on this PR.

One minor note (non-blocking): POSIX-sourced GUI as python -m emrg.gui would not match the regex — not a real invocation path (GUI launches via Electron / EMRG.app / EMRG.exe), so acceptable.

[argszero]

✅ LGTM — cycle (R1950)

Second independent-cycle review. Re-verified the working state of the change:

• The matcher (_match_emrg_client) correctly anchors module boundaries (-m\s+emrg(\.server)?(\s|$)), so lookalike module names (emrg.serverless, emrgx, emrg.gui) are rejected — covered by negative tests.
• _scan_emrg_client_pids excludes the caller's own pid and tolerates malformed ps lines.
• _stop_pids SIGTERM → 3s grace → SIGKILL leaves no stragglers; POSIX scan runs after daemon stop, catching daemons respawned by still-alive TUIs in the race window.
• stop-emrg.cmd step [0] is best-effort: old launchers (no stop subcommand) fail non-fatally and fall through to the robust steps [1]-[4]; first-time installs without the command continue cleanly.
• CI: test + test-windows both pass on this PR; mergeable CLEAN.

No blocking issues.

[argszero]

✅ LGTM — cycle

Reviewed the full diff fresh this cycle:

• _match_emrg_client — the -m\s+emrg(\.server)?(\s|$) regex is correctly anchored: whitespace-or-end after emrg/(.server) rejects lookalikes (emrg.serverless, emrgx, emrgistry), and the EMRG.app substring covers the macOS GUI. Negative-state tests cover module-like flags (-X dev) so no false positives.
• _scan_emrg_client_pids — parses ps -axww -o pid=,command=, excludes own pid so emrg stop never kills itself; scanning after _stop_daemon() also catches a daemon a still-alive TUI respawned in the race window.
• _stop_windows_clients — mirrors stop-emrg.cmd steps [1]/[2]: graceful GUI stop then unconditional /F fallback (host 2026-08-10T01:27:07Z lesson), TUI via PowerShell CIM command-line filter (wmic-free, Win11 24H2+ safe, excludes emrg.server).
• stop-emrg.cmd step [0] — best-effort: installed launcher preferred, PATH fallback, missing command → continue without error (first-time install); an old emrg.cmd stop (no subcommand) exiting non-zero is not gated on, so no version-safety regression.
• Tests — matcher positive+negative, scan parse/exclusion, step [0] ordering + no-%VAR% block guard, CRLF guard; doc count synced; CI test + test-windows both green.

No blocking issues.