feat(desktop): add pricing override management UI

[liuxiaocs7]

Summary

The Usage → Pricing tab was display-only. The Runtime Host pricing authority (pricing.query / pricing.mutate, PricingStore) and the Desktop IPC handlers (usage:pricing:{list,put,reset}) were already implemented and tested, but the preload never bridged the mutation channels and UsagePricingPanel had no add/edit/reset affordance — so a user could not register a custom per-model price, and a model missing built-in pricing reads as "cost unavailable".

This wires up the existing backend to a UI:

• Preload bridge — expose settings.pricing.put(config) / settings.pricing.reset(modelKey), scoped to the selected Runtime Host, unwrapping the Result so the form can surface a failure inline.
• Data shape — carry modelKey and optional cache rates on UsageStats.pricing so a row can be edited/reset by its canonical key; project them in the main process.
• UI — rework UsagePricingPanel: an "Add pricing" toolbar action (always visible, so it is discoverable from the empty state), per-row edit/reset, a Dialog form (provider, model, input/output rates, and a collapsible optional cache-rate section), and a destructive-confirm AlertDialog for reset. Rates are validated inline and again on the Host via normalizePricingConfig. After a mutation the panel reloads so the row and tab count update.

zh + en copy added; provider/model are locked while editing since they form the identity key.

Fixes #4163

Verification

Ran locally:

• tsc --noEmit for all four desktop projects (preload, main, renderer, storybook) — pass.
• @maka/core build — pass.
• node --test on runtime-host-usage-ipc-main.test.js (5), desktop-session-projection.test.js + runtime-host-client-pricing.test.js (22) — pass. The pricing-projection test asserts the new modelKey + cache-rate fields and fails without the projection change.
• biome check (lint + format) on all changed TS files — clean.

Not run: I did not launch the Electron app, so there is no screenshot/recording of the new UI yet. The renderer/preload wiring is covered by typecheck + lint against the shared bridge contract rather than an automated UI test. Happy to attach a screenshot on request.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code (Claude Opus 4.8) — investigated the existing pricing stack, implemented the preload bridge + panel UI + copy, and updated the projection test. Reviewed and submitted by the human contributor of record. Affected commit carries a Generated-by: Claude Code trailer.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No