[codex] fix coding agent model switching

[EKKOLearnAI]

Summary

Fixes coding agent session model switching so Claude/Codex sessions can switch provider/model without carrying stale runtime configuration from the previous provider.

Changes

• Enables model switching for coding agent sessions and adds a second protocol selection step before applying the switch.
• Persists default workspace paths for Hermes bridge sessions and preserves stored coding agent workspaces when switching models.
• Avoids notifying the Hermes bridge for coding agent model switches and clears stale native session ids when model/provider changes.
• Prevents stale baseUrl/apiKey from being reused after coding agent provider changes, with a backend guard for mismatched builtin provider endpoints.
• Uses stored session model/provider when deciding whether an existing coding agent run is still compatible, so stale runs are restarted with the new configuration.

Root Cause

Coding agent sessions stored provider runtime fields such as baseUrl and apiKey in frontend session memory. After switching provider/model, continuing without a page refresh could still send the previous provider endpoint, for example provider=deepseek with a Xiaomi base URL. The backend then generated a new Codex home for DeepSeek but registered the proxy against the stale upstream endpoint.

Validation

• npm exec -- vitest run tests/client/chat-store-reasoning-tool-boundary.test.ts tests/client/chat-panel-session-click.test.ts tests/server/coding-agent-resume-config.test.ts tests/server/handle-coding-agent-run.test.ts tests/server/sessions-controller.test.ts
• npm exec -- vitest run tests/client/chat-panel-session-click.test.ts tests/server/sessions-controller.test.ts tests/server/coding-agent-resume-config.test.ts tests/server/run-chat-bridge-final-context.test.ts
• npm run harness:check
• npm run build