feat: project-less scratch sessions

[AgentWrapper]

Implements #2803.

Adds a built-in Scratch pseudo-project so users can spawn agent sessions without registering a repo first.

Backend

• New domain constants: ProjectKindScratch and ScratchProjectID.
• Session service defaults empty projectId to scratch and falls back to the daemon's configured default harness.
• Session manager creates a throwaway bare repo under ~/.ao/scratch// per scratch session; existing gitworktree/branch machinery is reused unchanged.
• Project service pins Scratch at the top of List results.
• Spawn controller normalizes empty projectId to scratch.
• gitworktree Create now honors cfg.RepoPath so scratch sessions can supply their per-session repo.

CLI

• ao spawn falls back to Scratch when no project context is available.
• Scratch spawns with no explicit agent skip local preflight; harness resolution is deferred to the daemon.
• CWD project resolution skips the Scratch pseudo-project.

Tests

• New scratch spawn coverage in service/session, session_manager, and CLI.
• New project service test for the pinned Scratch summary.
• Updated HTTP controller project list/delete assertions to account for the pinned Scratch entry.
• Spawn tests are isolated from inherited AO_PROJECT_ID/AO_SESSION_ID.

Verification

• go test ./... passes in backend.
• golangci-lint reports no new issues in changed packages.
• OpenAPI spec/schema.ts unchanged (ProjectKind is an open string, no enum drift).
• Frontend typecheck could not run because node_modules is not installed in this worktree; this is a pre-existing environment issue unrelated to the change.

Closes #2803

[AgentWrapper]

Review: changes requested

The feature shape is good — reusing the existing gitworktree machinery by threading cfg.RepoPath through is the right call, and the spawn-path test coverage is solid. But there is one blocking lifecycle bug: the per-session repo path is never persisted, so every workspace operation after spawn (kill, shutdown save, restore, even in-spawn rollback) fails to resolve the repo for scratch sessions. Details in the inline comments; summary below.

Blocker

1. Scratch sessions cannot be killed, saved on shutdown, or restored. Spawn passes the bare-repo path to Workspace.Create via cfg.RepoPath, but nothing persists it. workspaceInfo(rec) (manager.go:2569) builds ports.WorkspaceInfo without RepoPath, so the adapter falls back to repoPath(info.ProjectID) and the daemon-wired projectRepoResolver errors with no project registered with id "scratch". Consequences:

   • Kill destroys the runtime, then workspace.Destroy errors, so Kill returns before MarkTerminated — a zombie session (dead runtime, non-terminated, un-removable from the dashboard).
   • SaveAndTeardownAll/boot reconcile: StashUncommitted errors, so scratch sessions are never shutdown-saved.
   • RestoreAll builds its WorkspaceConfig without RepoPath, so restore fails.
   • In-spawn rollback (destroySpawnWorkspace) fails too, since Workspace.Create returns info without RepoPath (workspace.go:145).

   The new manager test only exercises Spawn, which is why this went unnoticed. Persist the repo path (e.g. in SessionMetadata or the session_worktrees row) and thread it through workspaceInfo/restore config — or derive <dataDir>/scratch/<sessionID> for scratch sessions in those paths.

Should fix

2. Reserved id not enforced: validateProjectID still accepts scratch, so ao project add --project-id scratch succeeds; the list then has two scratch entries and the registered project is unreachable (spawns silently land in the throwaway repo).
3. List/Get inconsistency: GET /api/v1/projects includes scratch but GET /api/v1/projects/scratch returns 404. The CLI already special-cases around this; other consumers (frontend picker) will hit the 404.
4. Disk leak: nothing ever removes <dataDir>/scratch/<sessionID>; every throwaway session permanently leaks a bare repo under ~/.ao/scratch.

Minor

5. The CLI now silently falls back to Scratch where it previously errored with guidance — a user spawning from inside an unregistered repo may expect the agent to work on that repo. A one-line notice would help.
6. The hard-coded SHA-1 empty-tree oid breaks under init.defaultObjectFormat = sha256.

Verified locally: backend builds and the new tests in cli, service/session, service/project, and session_manager all pass.

[AgentWrapper]

Follow-up to review 4729425711 — all six points addressed in 39f15c8 (rebased onto main; the rebase also resolved the merge conflicts with #2850's --name change):

• Blocker (persist repo path): nothing persisted; the path is derived deterministically as /scratch/ on every post-spawn workspace path — Kill, SaveAndTeardownAll, Cleanup, RestoreAll/restoreSessionWorkspace, and spawn rollback (Create now returns RepoPath).
• Disk leak: the throwaway bare repo is removed after a successful destroy on kill/cleanup/rollback.
• sha256: empty-tree oid now comes from git mktree in the target repo.
• Reserved id: Add and Remove both reject 'scratch' with PROJECT_SCRATCH_RESERVED.
• List/Get inconsistency: GET /api/v1/projects/scratch now returns the pseudo-project (kind presented as 'scratch').
• CLI fallback notice: spawn prints a one-line note when it falls back to Scratch.

Additionally, end-to-end verification against the real SQLite store surfaced a bug my fake-store tests had masked: the sessions.project_id FK rejected scratch session inserts. Fixed by seeding the built-in scratch project row via migration 0025 (kind stored as 'single_repo' because the projects.kind CHECK predates scratch; the service layer always presents it as 'scratch'). Store-level regression test included.

Backend: go build ./... and go test ./... green after rebase. Live verification on a dev daemon: zero-setup spawn (no --project/--agent) works, GET projects lists Scratch pinned first, GET projects/scratch → 200, kill → freed:true + terminated + repo dir removed.