[Feat] Enable environment and task snapshot resume on Daytona

[roomote-roomote]

Opened on behalf of Matt Rubens. Follow up by mentioning @openmote, in the web UI, or in Slack.

Related issue

N/A — internal Roomote work requested in Slack after reviewing Daytona pause/fork / Snapshot product gaps.

Why this PR exists

• A maintainer explicitly invited this PR in the linked issue or discussion
• I am a maintainer / this is internal Roomote work

What changed

Enable Daytona for Roomote’s product snapshot flows (environment setup snapshots and task sleep/resume), using durable named filesystem snapshots like Modal and E2B rather than pause-in-place.

• Implement DaytonaClient.createSnapshot via _experimental_createSnapshot (name as snapshotId, destroy source sandbox afterward)
• Implement resumeFromSnapshot by creating a sandbox from that named snapshot
• Extend createDaytonaMachine / spawn-daytona-worker for fresh / environment_snapshot / task_snapshot launch modes (mirrors E2B)
• Mark Daytona snapshot/resume capable in types + setup catalog so sleep-check auto-snapshots resumable runs
• Keep DAYTONA_SNAPSHOT_NAME as the worker base image only; product snaps use roomote-run-snap-*
• Docs distinguish worker base snapshots vs product snapshots

How it was tested

• pnpm --filter @roomote/compute-providers exec vitest run src/__tests__/adapters.contract.test.ts (create + resume + destroy after snapshot)
• pnpm lint && pnpm check-types
• pnpm --filter @roomote/docs check
• No live Daytona API spike in this environment (experimental API path is unit-tested with mocks)

Checklist

• The PR title follows the repo convention: [Fix], [Feat], [Improve], [Refactor], [Docs], or [Chore] followed by a user-facing description
• This PR is small and scoped to one change
• pnpm lint and pnpm check-types pass locally
• I added tests or included a clear manual validation note above
• I removed secrets, tokens, private keys, and customer data from code, logs, and screenshots
• If this change should appear in the changelog, I ran pnpm changeset

[roomote-roomote]

No new code issues found. The earlier comment nit is now fixed. See task

• apps/bullmq/src/scheduled-jobs/sleep-check.ts:156-159 — Stale comment: the isSnapshotResumableSleepCandidate JSDoc still said "Non-snapshot providers (Daytona) always fall through to the destroy paths." Fixed in 909ce31: the JSDoc now cites Docker as the non-snapshot example and lists Modal, E2B, and Daytona as snapshot-capable, matching snapshotCapableComputeProviders.

Summary

This enables Daytona for Roomote's snapshot/resume flows by porting the E2B model (durable named filesystem snapshots + fresh/environment_snapshot/task_snapshot launch modes). I verified:

• spawnDaytonaWorker is essentially line-for-line equivalent to spawnE2bWorker, including launch-mode selection, the SnapshotResume missing-sourceSnapshotId guard, and worker command/args.
• createDaytonaMachine's shouldInstallShippedRuntime = launchMode !== 'task_snapshot' and the task_snapshot early-return match createE2bMachine.
• _experimental_createSnapshot(name, timeout?) and getPreviewLink(port) match the installed @daytonaio/sdk@0.190.1 types.
• createSnapshot destroying the source sandbox after capture matches Modal's snapshot-kills-sandbox contract, and the snapshot BullMQ job + sleep-check already build a Daytona client whose snapshotName resolves from DAYTONA_SNAPSHOT_NAME.
• reconcileSnapshottingFailure gracefully handles Daytona not implementing the optional findSnapshotBySourceInstance.
• snapshotCapableComputeProviders now includes daytona, and sleepCheckManagedComputeProviders (spread) resolves to the same {modal, e2b, daytona} set as before.

Sync update (head 909ce31): The only change since the prior review is the docs fix above, which resolves the single outstanding checklist item. No new code issues.

Non-blocking note (low confidence): Because createSnapshot derives a new random snapshot name per call, destroys the source sandbox, and Daytona has no findSnapshotBySourceInstance reconcile, each sleep/resume cycle leaves the previous named snapshot orphaned, and a partial failure after a snapshot completes server-side but before DB persistence cannot be recovered (the run is marked failed instead). This mirrors accepted cross-provider snapshot behavior and DB-level expiry (SANDBOX_SNAPSHOT_EXPIRY_MS), so it may be intentional — worth confirming whether orphaned Daytona named snapshots get garbage-collected.