feat(dashboard): show weekly token pace

[Komzpa]

Summary

• add a dashboard Weekly pace card that compares actual weekly token consumption against schedule
• compute pace by summing token/credit budgets per account, using each account's own reset time and window
• model weekly reset replenishment as the full account budget, not only current unused credits
• hide Recovery options when the pool is on pace and no throttle/account action is needed
• preserve weekly capacity/remaining fields in the frontend schema and document the behavior in OpenSpec

Why

A nearly empty account near its weekly reset should not look scary, and a healthy pool should not tell the operator to do recovery work. The card compares actual remaining tokens to expected remaining tokens per account, then derives display percentages from aggregate token totals instead of averaging account percentages.

Live data sanity check

Checked the real codex-lb usage history around the "tokens were insufficient, then an account was added, then tokens were sufficient" case:

• 2026-05-03 18:13 UTC: 6-account pool was danger, about 91.5% used vs 76.4% scheduled, short by about 18.8K credits.
• 2026-05-03 18:16 UTC: after adding the unused Pro account, actual 7-account pool became on track, about 64.1% used vs 53.6% scheduled, projected low-water about 31.7K credits.
• Counterfactual without the new Pro account at 2026-05-03 18:16 UTC stayed danger, short by about 18.7K credits.

That matches the intended continuity context: adding a real account turns the pool from insufficient to sufficient, and the card should stop showing Recovery options once the pool is safe.

Validation

• cd frontend && bun run test -- ./src/features/dashboard/utils.test.ts ./src/features/dashboard/components/weekly-credits-pace-card.test.tsx
• cd frontend && bun run typecheck
• cd frontend && bun run lint (passes with two pre-existing warnings in account-multi-select.tsx)
• cd frontend && bun run build (passes with existing Vite chunk-size/dynamic-import warnings)
• git diff --check
• earlier branch validation also covered .venv/bin/python -m pytest tests/integration/test_dashboard_overview.py::test_dashboard_overview_combines_data

openspec CLI is not installed in this workspace, so I could not run local OpenSpec validation.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5df035489a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Model first reset as full replenishment

In buildWeeklyPoolProjection, the first reset event initializes topUpCredits as fullCredits - remainingCredits, which is the amount already consumed, not the amount needed at reset time to restore the account to full. This underestimates post-reset capacity and can falsely project shortfall/depletion (and therefore trigger "danger" status plus pause/throttle recommendations) for accounts whose real burn is sustainable across weekly resets; for example, an account mid-cycle with moderate usage can be marked as depleting even though each reset would keep it above zero.

Useful? React with 👍 / 👎.