fix(frontend/library): autopilot pill counts copilot follow-ups only, not graph schedules (#13212)

## Why

Spotted in the screenshot below — the autopilot library pill on
`/library` was double-counting:

- The briefing's "**Scheduled 3**" tab already counts agents that have
schedules attached.
- My PR #13202 added an "**Autopilot library — 4 scheduled**" pill at
the bottom that sums copilot follow-ups + graph schedules (1 + 3 = 4).
- Reads as inconsistent ("3 vs 4") and dilutes the autopilot framing —
graph schedules aren't an autopilot/copilot concept, they're an agent
concept covered by the tab above.

Also: with the previous behaviour, a user who's never touched the
copilot but has 1 graph schedule would still see an "Autopilot library 1
scheduled" pill, even though they have no autopilot-specific content.

## What

- Drop graph-schedules query + count from the pill.
- Count copilot follow-ups only.
- Rename link label `X scheduled` → `X follow-up(s)` so the count
clearly belongs to autopilot.
- Pill still hides entirely when skills=0 AND follow-ups=0 (graph
schedules alone no longer surface the autopilot header — they belong to
the "Scheduled" briefing tab).
- Link destination unchanged — still `/library/followups`, which renders
BOTH copilot follow-ups AND graph schedules (the unification from #13202
stays useful at the destination).

## How

```diff
-import {
-  useGetV1ListExecutionSchedulesForAUser,
-  useListCopilotFollowupSchedules,
-} from "@/app/api/__generated__/endpoints/schedules/schedules";
+import { useListCopilotFollowupSchedules } from "@/app/api/__generated__/endpoints/schedules/schedules";
@@
-  const { data: graphsRes } = useGetV1ListExecutionSchedulesForAUser({…});
-  …
-  const scheduledCount = copilotFollowupsCount + graphSchedulesCount;
-  if (skillsCount === 0 && scheduledCount === 0) return null;
+  const followupCount = …;
+  if (skillsCount === 0 && followupCount === 0) return null;
@@
-  {scheduledCount} scheduled
+  {followupCount} follow-up{followupCount === 1 ? "" : "s"}
```

No new feature flag — count-based hide already keeps the pill quiet for
first-time users.

## Test coverage

Updated the BriefingTabContent vitest cases:

- `counts copilot follow-ups only (graph schedules are NOT folded in —
they have their own briefing tab)` — asserts graph schedules don't
inflate the count.
- `hides the pill entirely when skills=0 AND copilot follow-ups=0 (even
if graph schedules exist)` — asserts graph schedules alone don't surface
the autopilot label.
- `shows both links with a separator when skills AND follow-ups are
positive (singular pluralization)` — verifies `1 skill` / `1 follow-up`
singulars.

All 24 BriefingTabContent + 17 CostsBreakdown tests pass. Types clean.

## Out of scope

- No backend changes.
- Not changing the unified `/library/followups` destination page — it
correctly renders both kinds.

## Checklist

- [x] Code follows project style guidelines
- [x] Self-review performed
- [x] Tests added for new behaviour (graph-schedules-don't-count +
hide-on-zero with graph schedules present)
- [x] No new warnings
- [ ] CI green