feat: add per-model context and thinking editor mirroring the Qoder IDE

[luoxuanzao]

Summary

• What changed?
  • Model selector dropdown becomes an IDE-style two-card cascade: the model list stays in the left pane while a per-model editor card expands beside the hovered row (row-anchored, with flip/max-height placement extracted as pure, tested functions).
  • The editor shows context window tiers (sorted ascending), a thinking on/off toggle, and the model's server-provided thinking effort levels sorted by the canonical intensity scale (low → medium → high → xhigh → max), matching the Qoder IDE selector.
  • Per-model overrides (contextWindow / thinkingEffort / thinkingEnabled) persist in settings and apply to every request via SDK pull-mode resolveModel; probe maps context_config / thinking_config, and getters normalize stale persisted order on read.
  • Hardening from self-review: editing a non-selected model no longer rewrites the conversation usage meter; reopening the dropdown (mouse or keyboard) returns to the list view; runtime degradation clears a stale editor card; edit pencil is hover/focus-only.
  • i18n keys added in 10 locales; CHANGELOG entry added.
• Why is it needed?
  • Brings Qoderian's model configuration to parity with the Qoder IDE: users can tune context window and thinking effort per model instead of a single global effort level.

Verification

• npm run typecheck
• npm run lint
• npm run test (3301 tests, 141 suites)
• npm run build
• npm run release:check
• npm run audit:prod (0 vulnerabilities)
• Tested affected qodercli behavior against a real CLI when applicable (live catalog probe + end-to-end checks in a real Obsidian vault: tier/effort ordering, persistence, override clear semantics, usage meter, reopen/degrade cleanup)

Safety

• No credentials, private vault content, internal URLs, or personal paths are included
• User-visible changes are documented in CHANGELOG.md