Usage API now reports per-model weekly limits in a `limits` array — widget misses the binding cap (e.g. Fable weekly)

[dmriding]

The /api/oauth/usage endpoint now includes a limits array, and per-model weekly caps only appear there — the legacy seven_day_opus / seven_day_sonnet fields are always null now. Since the monitor parses only five_hour and seven_day, users on plans with a model-scoped weekly cap can hit rate limiting while the widget shows a comfortable all-models weekly.

Real example from my account today: all-models weekly at 26%, but Fable weekly at 43% with is_active: true — the binding limit was invisible in the widget.

Trimmed real response:

{
  "five_hour":  { "utilization": 29.0, "resets_at": "..." },
  "seven_day":  { "utilization": 26.0, "resets_at": "..." },
  "seven_day_opus": null,
  "seven_day_sonnet": null,
  "limits": [
    { "kind": "session",       "percent": 29, "resets_at": "...", "scope": null, "is_active": false },
    { "kind": "weekly_all",    "percent": 26, "resets_at": "...", "scope": null, "is_active": false },
    { "kind": "weekly_scoped", "percent": 43, "resets_at": "...",
      "scope": { "model": { "id": null, "display_name": "Fable" } }, "is_active": true }
  ]
}

Suggested handling: parse limits[] entries with kind: "weekly_scoped", take the label from scope.model.display_name, and treat is_active as "this is the currently binding limit".

I've implemented this in my own (heavily diverged, private) fork and it works well — happy to send a PR adapted to your codebase if you'd like one. Thanks for building this!