Green up master: horizon balance test drift + roster query-budget N+1

You are being dispatched from flower Brief #23: Green up master: horizon balance test drift + roster query-budget N+1

Recall pointer:
- Use recall_brief with id 23 for the full folder if you need provenance.

Target:
- project: flower (/Users/mikeferrara/Documents/code/flower)
- branch: flower/greenup-test-debt
- worktree: not specified
- kind: fresh

Current brief spec:
## Goal
Green up `master` — two baseline `php artisan test` failures inherited from this session's
merged changes. Fix BOTH on one branch.

## Failure 1 — QueueTimeoutAlignmentTest (stale expectation)
`tests/Feature/QueueTimeoutAlignmentTest.php::test_horizon_supervisors_split_fast_and_long_work`
asserts supervisor-long `balance === 'simple'`, but the long/summarize supervisor was
intentionally changed to **`balance => 'auto'`** (autoscale min 1 → max 12; commit 8c60cd5,
config/horizon.php). The config is correct; the test is stale.
- Update the expectation to `'auto'`. Keep the rest of the assertion meaningful (fast vs long
  split still enforced; long supervisor still isolated with its generous timeouts / minProcesses).

## Failure 2 — QueryEfficiencyAuditTest roster budget (N+1)
`tests/Feature/QueryEfficiencyAuditTest.php::test_key_pages_have_bounded_query_counts` — the
/roster page now runs 6 queries (budget 4). Root: `DaemonRosterService::linkedSessionLastActivity()`
(added in 28ddf12 for session-link liveness) runs **one `sessions` query PER daemon** from the
Roster/Index each-loop (`select last_activity_at from sessions where external_session_id = ? ...`).
- **Fix by BATCHING, not by bumping the budget** — it must stay O(1) in query count regardless of
  daemon count. Load all rostered daemons' linked-session `last_activity_at` in a SINGLE query keyed
  by `external_session_id` (e.g. `Session::whereIn('external_session_id', $ids)->groupBy(...)->selectRaw(max last_activity_at)`),
  build a map once, and have `livenessFor()` / `effectiveLastActivity()` read from that preloaded map
  instead of per-daemon queries. Keep the derived liveness result byte-identical to today
  (alive/stale/dead thresholds unchanged; the 'alive'→'live' UI mapping unchanged).
- Keep the /roster page under the 4-query budget with the seeded multi-daemon fixture.

## Acceptance
- `QueueTimeoutAlignmentTest` and `QueryEfficiencyAuditTest` both pass.
- Roster liveness behavior unchanged (existing Roster/Index tests still green).
- `php artisan test` targeted classes green; pint clean on changed files.
- `Brief: #<id>` trailer on every commit; brief_append when done.

## Provenance
Debt discovered 2026-07-01 while reviewing Brief #21 — the backend worker flagged these two
baseline failures. Both trace to this session's merges (autoscale 8c60cd5, roster liveness 28ddf12).

Recent/key trace events:
[1] participant_joined flower-orchestrator: (no body)
[2] note_added flower-orchestrator: Two baseline test failures inherited from this session (autoscale 8c60cd5 + roster liveness 28ddf12). Fix QueueTimeoutAlignmentTest expectation (auto) + batch the roster session-activity lookup to stay under the query budget.
[3] plan_proposed flower-orchestrator: ## Goal
Green up `master` — two baseline `php artisan test` failures inherited from this session's
merged changes. Fix BOTH on one branch.

## Failure 1 — QueueTimeoutAlignmentTest (stale expectation)
`tests/Feature/QueueTimeoutAlignmentTest.php::test_horizon_supervisors_split_fast_and_long_work`
asserts supervisor-long `balance === 'simple'`, but the long/summarize supervisor was
intentionally changed to **`balance => 'auto'`** (autoscale min 1 → max 12; commit 8c60cd5,
config/horizon.php). The config is correct; the test is stale.
- Update the expectation to `'auto'`. Keep the rest of the assertion meaningful (fast vs long
  split still enforced; long supervisor still isolated with its generous timeouts / minProcesses).

## Failure 2 — QueryEfficiencyAuditTest roster budget (N+1)
`tests/Feature/QueryEfficiencyAuditTest.php::test_key_pages_have_bounded_query_counts` — the
/roster page now runs 6 queries (budget 4). Root: `DaemonRosterService::linkedSessionLastActivity()`
(added in 28ddf12 for session-link liveness) runs **one `sessions` query PER daemon** from the
Roster/Index each-loop (`select last_activity_at from sessions where external_session_id = ? ...`).
- **Fix by BATCHING, not by bumping the budget** — it must stay O(1) in query count regardless of
  daemon count. Load all rostered daemons' linked-session `last_activity_at` in a SINGLE query keyed
  by `external_session_id` (e.g. `Session::whereIn('external_session_id', $ids)->groupBy(...)->selectRaw(max last_activity_at)`),
  build a map once, and have `livenessFor()` / `effectiveLastActivity()` read from that preloaded map
  instead of per-daemon queries. Keep the derived liveness result byte-identical to today
  (alive/stale/dead thresholds unchanged; the 'alive'→'live' UI mapping unchanged).
- Keep the /roster page under the 4-query budget with the seeded multi-daemon fixture.

## Acceptance
- `QueueTimeoutAlignmentTest` and `QueryEfficiencyAuditTest` both pass.
- Roster liveness behavior unchanged (existing Roster/Index tests still green).
- `php artisan test` targeted classes green; pint clean on changed files.
- `Brief: #<id>` trailer on every commit; brief_append when done.

## Provenance
Debt discovered 2026-07-01 while reviewing Brief #21 — the backend worker flagged these two
baseline failures. Both trace to this session's merges (autoscale 8c60cd5, roster liveness 28ddf12).
[4] status_change flower-orchestrator: (no body)

Recommended linked context:
{
    "todos": [],
    "scratchpads": []
}

Execution notes:
- Treat the brief as the source of truth.
- Keep work scoped to this dispatch request.
- Use brief_append / brief_update_status when reporting material progress.
- Add a git commit trailer `Brief: #23` to every commit for this brief so flower can exact-link commits back to the brief.