[Enhancement] Settings: expose actual Codex model selection and per-call/per-tick cost breakdown by caller

[Peefy]

Is your feature request related to a problem? Please describe.
The user picks Codex as the model provider but cannot tell which specific model OpenLoomi is calling, where the call originates (chat vs tick vs workflow), or how much each call costs. The token total in settings aggregates across all callers, so Loop ticks and one-off chat replies look the same.

Describe the solution you'd like

• Settings → Models: show the resolved model id per provider, plus the last 24h token/cost breakdown by caller (chat / Loop tick / Loop brief / Loop wrap / memory workflow / standalone skill).
• Allow pinning a model per caller class (e.g. "Loop ticks use gpt-4o-mini, chat uses gpt-4o").
• Show the Codex billing/quota status if the provider exposes it.

Describe alternatives you've considered

• Rely on Codex's own dashboard. Rejected: doesn't break down by OpenLoomi caller.
• A single global model dropdown (status quo). Rejected: coarse-grained; the per-caller breakdown is the value.

Additional context
Multiple call sites share the same model client; instrumenting the call to thread caller: 'loop.tick' | 'chat' | ... is the prerequisite. Once caller is threaded, the existing per-tick summary record (see #9) can aggregate naturally.