fix(session): make worker branch namespace child-safe

[neversettle17-101]

Fixes #308

Summary

• Change default worker branches from ao/<session-id> to ao/<session-id>/root so sibling PR branches like ao/<session-id>/<topic> are valid Git refs.
• Teach SCM PR attribution that a /root session branch also owns sibling branches under the same namespace.
• Update worker branch instructions and CLI help text.

Root Cause

AO created the session root branch as ao/<session-id> while instructing agents to create additional PR branches as ao/<session-id>/<topic>. Git cannot create both refs because refs/heads/ao/<session-id> blocks refs/heads/ao/<session-id>/<topic>.

Tests

• GOCACHE=/private/tmp/ao-go-build-cache go test ./internal/session_manager
• GOCACHE=/private/tmp/ao-go-build-cache go test ./internal/observe/scm
• GOCACHE=/private/tmp/ao-go-build-cache go test ./internal/cli -run 'TestSpawn'

[Pritom14]

@neversettle17-101 can you check this edge case :

If two sessions in the same repo somehow end up with ao/p-1 (no suffix) AND ao/p-1/root, both claim the prefix ao/p-1 for sibling PRs and the longest-prefix tiebreak ties at length 6. The winner is then determined by iteration order of byRepo[repoKey] — nondeterministic. Realistic only if a user manually picks ao/<id> as a branch override while another defaulted session lives at ao/<id>/root.

[neversettle17-101]

On the ao/p-1 vs ao/p-1/root edge case: I do not think this will show up in many normal flows.

The ambiguous case requires all of these to exist in the same repo:

• Session A stored branch: ao/p-1
• Session B stored branch: ao/p-1/root
• PR head branch: ao/p-1/something

That is unlikely because existing sessions from before this fix may have ao/p-1, but newly spawned sessions after the fix will get new IDs like ao/p-2/root, ao/p-3/root, etc. AO should not naturally create both ao/p-1 and ao/p-1/root for two different sessions. Also, if ao/p-1 still exists as a Git ref, creating ao/p-1/root usually fails at Git ref creation anyway.

The more realistic compatibility case is an old running session with session.branch = ao/p-1 trying to create ao/p-1/topic; Git will reject that because ao/p-1 already exists as a leaf ref. That affects only sessions spawned in the short window after multi-PR support landed and before this fix. Those sessions can finish their root PR, or any fallback PR can be manually claimed if needed.

So I would keep this PR simple and not add extra tie-break logic unless we see this ambiguity in real state or a reviewer wants the deterministic fallback explicitly.