fix(components): keep presence snapshots across bounded reconnect stop

[slashdevcorpse]

Related issue

Closes #480

Problem / pressure

#449 bounded renderer meta reconnect backoff. Each retry still stops the presence transport, which published {}. The UI then treated a live PC as offline. That is the renderer side of the same cluster as #398 / #399.

Summary

WorkspacePresenceTransport no longer publishes a snapshot with no kind: 'machine' entries. stop() does not emit {}. A reconnect start() on an empty store does not either. Wake detection still stamps lastSnapshotAtMs. Last machine heartbeats remain until a new machine snapshot or the 90s TTL.

Before / after

Before	After
stop() or the next empty start() emitted {} and lists showed the PC offline.	Those paths keep the last machine snapshot. Sync state is idle.
#449 bounded retry rate only.	Bounded retries no longer wipe liveness.

Test plan

• corepack pnpm exec vitest run tests/workspace-presence-transport.test.ts tests/local-reconnect-loop.test.ts tests/create-workspace-runtime-meta-recovery.test.ts from packages/components — 34 passed.
• The presence-stop test failed first ({} vs last machine snapshot) and passed after the skip.
• fix(components): bound meta reconnect recovery #449 reconnect-loop tests still pass.
• No live desktop reconnect session was run.

Context handoff

Instructions for reviewing agents

• Review focus: onBeforeStop and onStoreChange in workspace-presence-transport.ts, plus the keep-on-stop / restart-empty case in workspace-presence-transport.test.ts.
• Decisions to challenge: Skip any snapshot with no kind: 'machine' instead of a UI .has() sweep. Viewing-only updates no longer notify onSnapshot.
• Plausible failures / evidence gaps: A machine that truly leaves can look online until TTL (90s). Viewing presence in the UI snapshot can lag across reconnect until the next machine heartbeat. Workspace switch still clears via clearLodyPresenceStatesAtom.

Authoring context

• User goal / directives: File a brief bug that cites fix(components): bound meta reconnect recovery #449 and open a matching fix PR for the remaining false-offline wipe during bounded renderer reconnect.
• Constraints / non-goals: Do not retune fix(components): bound meta reconnect recovery #449 backoff. Do not change CLI watchdog ([Bug] Streams reconnect loop: watchdog tears down a connected transport because its meta room is still joining #399 / fix(cli): skip Streams reconnect while a meta room join is in flight #478), Flock freshness ([Bug] Session create hard-fails on a best-effort Flock freshness sync when the network is degraded #398), or MCP error codes ([Bug] MCP error classification: network failures marked non-retryable and unactionable operationId retry advice #400). Do not sweep every UI .has() call site.
• Risk-bearing decisions: Stale cloud presence can linger up to the 90s TTL after stop. Local-origin merge is unchanged.
• Destructive or irreversible behavior: None. Stop still tears the room down; it only stops publishing an empty map.
• Deliberately not done or tested: No live Electron reconnect. No MCP MACHINE_OFFLINE change. No three-state UI sweep.
• Unknowns / confidence: High on the transport unit test. Medium that every consumer reads the last snapshot rather than treating idle as empty.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

@slashdevcorpse, this pull request needs updates before review.

It is marked status:needs-pr-attention. Address the findings below by 2026-09-16 08:24:01 UTC. The label and this comment are removed automatically after the PR passes validation.

If the PR remains invalid for 7 days, it will be closed and marked status:pr-policy-expired. Continue afterward by opening a new pull request with the current template.

Policy findings

PR does not meet Lody contribution requirements:

- Missing required heading: ## Visual explanation

See `CONTRIBUTING.md` and `.github/PULL_REQUEST_TEMPLATE.md`.