fix(frontend): clean up agent briefing cost breakdown UX (#13196)

## Summary

Addresses Toran's [Discord
feedback](https://discord.com/channels/1126875755960336515/1258404661048770590/1506972972655706212)
on the new costs breakdown:

- **Drops AutoPilot rate-limit meters** from the library briefing panel
— those belong with Copilot, not here.
- **Bars now scale to share of monthly spend** (not to the top spender)
with an explicit `%` label and a `% of monthly spend` caption. Solves
"what is this percentage measuring?"
- **Filters zero-cost agents** out of *Spend by agent* — no more `$0.00
· 9 runs` rows with a small bar.
- **Clarifies the period**: subtitle reads `Calendar month so far · May
1 – today (UTC)` (matches the backend default at
`execution_cost_summary.py:70`).
- **Per-agent avg / run pill** next to the run count.

## Files

- `BriefingTabContent.tsx` — removed `UsageSection` + `UsageMeter`, now
just renders `<CostsBreakdown />` on the `all` tab
- `CostsBreakdown.tsx` — period subtitle, passes `totalCents` to
`SpendByAgentList`
- `SpendByAgentList.tsx` — share-of-total scaling, `%` label, zero-cost
filter, avg-per-run pill
- `__tests__/BriefingTabContent.test.tsx` — drops obsolete usage-meter
tests, adds three new tests for the changes above

## Not addressed (separate decision)

Ubbe's note: when monthly spend is cents-only, the whole breakdown
collapses into `$0.00` / `$0.01` rows. That needs a product call (drop
the breakdown below a threshold? sub-cent precision in `formatCents`?).
Left for a follow-up.

## Test plan

- [ ] On the `all` tab, no `Usage limits` / `Today` / `This week` /
`Manage billing` UI
- [ ] *See costs breakdown* expands to show the new `Calendar month so
far · …` subtitle
- [ ] Bars and `%` labels in *Spend by agent* sum (roughly) to 100%
- [ ] Agents with $0.00 spend don't appear in *Spend by agent*
- [ ] Per-agent rows show `N runs · avg \$X.YZ`
- [ ] Aggregate *Avg / run* still divides by `billable_run_count`
(existing test asserts \$5.00, not \$1.67)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>