Fix session model persistence and default reasoning behavior

[roackb2]

Summary

• preserve the target session model when switching sessions in the TUI
• lower the default reasoning effort for gpt-5.4 and gpt-5.5 family models from high to medium
• add focused tests for session model sync and reasoning status reporting

Verification

• node_modules/.bin/vitest run src/__tests__/unit/tui/session-model-sync.test.ts src/__tests__/unit/tui/local-commands.test.ts src/__tests__/unit/core/openai-oauth.test.ts src/__tests__/unit/core/chat-turn-runtime.test.ts
• node_modules/.bin/vitest run src/__tests__/integration/tui/ask-cli.test.ts -t "runs a stateless ask|creates a new session|continues an existing session|preflight compacts"
• node_modules/.bin/vitest run src/__tests__/integration/chat/chat-runtime.test.ts -t "clears explicit reasoning effort when control-plane settings send null"
• yarn build
• yarn eslint src/cli/chat/App.tsx src/cli/chat/utils/session-model-sync.ts src/core/llm/model-policy.ts src/__tests__/unit/tui/session-model-sync.test.ts src/__tests__/unit/tui/local-commands.test.ts src/__tests__/unit/core/openai-oauth.test.ts

Notes

This does not yet centralize model and reasoning ownership. The next step should be a focused refactor around session execution preferences so new-session defaults, session persistence, ask mode, and UI state all share one owner.