feat(linux): add systemd --user discovery to Services.ts status/doc

[takanorinishida]

What's broken

Services.ts's own doc comment says status "reports reality, not a hand-maintained guess" — but loadedLabels()/findPlist()/cadenceOf() only ever look at launchctl/~/Library/LaunchAgents, so on Linux every one of the 16 registered services reports "✗ missing" regardless of what's actually installed. After the systemd backends for WorkSweep (#1692) and DerivedSync/CommitmentSweep/UsageAggregator/CodexUpdate (#1704) land, that's no longer true — but Services.ts has no way to know it.

Fix

Adds a process.platform branch to the three discovery functions:

• loadedLabels(): darwin parses launchctl list; linux parses systemctl --user list-units --all --type=service,timer,path, stripping the .service/.timer/.path suffix so a label counts as loaded from any of its unit files (same "loaded, not necessarily running" semantics launchctl list already had).
• findPlist(): darwin looks for an installed plist then a template in TOOLS/PULSE; linux looks for an installed unit (preferring .timer or .path over .service, since that's where the schedule lives) then a template, falling back to a bare .service for persistent daemons with no timer pairing (pulse — the systemd analog of a RunAtLoad-only launchd job) or a bare non-.template unit (pulse's own manage.sh keeps its systemd source that way, mirroring the plist side's existing dual lookup).
• cadenceOf(): darwin parses StartInterval/StartCalendarInterval/WatchPaths/RunAtLoad from XML; linux parses PathModified/OnCalendar/OnUnitActiveSec/Type=simple from the ini-style unit file.

install/uninstall need no branch — they just run each service's install: command, which is itself platform-dispatching inside the Install*.ts script; Services.ts never had to know how a service installs itself, only how to tell whether it did.

Services with no Linux port yet (com.lifeos.pulse-menubar, com.lifeos.deriver — both launchd-only by design, deriver explicitly per its own manage-deriver.sh: "Linux/systemd analog is intentionally not implemented") correctly continue to report "✗ missing" on Linux. That's fact, not a regression.

Why this doesn't change macOS behavior

Every deletion in this diff is a function signature line whose body moved unchanged into a renamed *Darwin variant, with the original name now the platform-dispatching wrapper — same pattern used across InstallWorkSweep.ts and the four Install*.ts scripts this depends on.

Verification

Tested on Ubuntu 24.04 / systemd 255 with all 6 currently-installed systemd services live (pulse, derivedsync, worksweep, commitmentsweep, usage-aggregator, codexupdate):

bun Services.ts status
  ● running    at load         Pulse (dashboard server)
  ● running    on file-change  Derived-file sync
  ● running    every 1h        Work sweep
  ● running    daily/scheduled Commitment sweep
  ● running    daily/scheduled Codex update
  ● running    daily/scheduled Usage aggregator

Every cadence string matches the actual installed unit content (OnUnitActiveSec=60min → "every 1h", OnCalendar daily entries → "daily/scheduled", derivedsync's .path PathModified → "on file-change", pulse's Type=simple with no timer/path → "at load"). The 10 not-yet-ported services (Conduit, healthsync, blogdiscovery, etc.) correctly show "✗ missing". bun Services.ts doc also verified — produces the same markdown table shape as the darwin-generated one, reflecting Linux's currently-narrower coverage.

darwin path unaffected: bunx tsc --noEmit shows only pre-existing @types/node-absence noise, no new errors; the 3 renamed *Darwin functions are byte-for-byte identical to their pre-change bodies.

Depends on: #1704 (the 4 installers whose systemd units this reads).

🤖 Generated with Claude Code