feat(settings): refine token usage and model configuration

[EF1874]

Summary

• add model filtering to token usage, refresh usage when settings regains focus, and resolve usage from the active Memmy home
• simplify the platform model catalog and show the platform Agent model as agent_chat from Memmy Platform
• expose distinct local and cloud platform Embedding choices with explicit assignment and account-owner validation

Validation

• npm run test -w @memmy/frontend-desktop — 160 files, 1530 tests passed
• changed Agent token-usage integration test — 6 tests passed
• Backend — 109 files, 834 tests passed with a local 10s Vitest timeout
• Memory — 75 files, 622 passed, 1 skipped
• frontend and Agent production builds passed
• frontend, Agent, Backend, and Memory typechecks passed
• Backend ESLint and git diff --check passed

Unrelated local full-suite observations

• @memmy/desktop full tests reported 5 failures in Windows process/upgrade-relay tests; none of those files are changed here
• App/memmy-agent full tests reported 28 failures across Windows path, Unix tool, Office archive, and process-timing tests; the changed token-usage integration test passes 6/6