fix(opencode): preserve the session model on resume

[devswha]

Summary

• prefer an explicit ChatMux model change, then the provider session's actual model, when resuming
• preserve OpenCode's providerID/id model identifier from its session database
• add regression coverage for both precedence and OpenCode model parsing

Root cause

The native OpenCode session used openai/gpt-5.5, but ChatMux resumed it with the new-session UI default anthropic/claude-sonnet-4-5. OpenCode stores the model as { providerID, id }; ChatMux also discarded providerID, so it could not reconstruct the CLI model argument correctly.

Verification

• targeted provider/OpenCode tests (15/15)
• npm run typecheck
• npm run lint
• npm run build

[devswha]

E2E QA evidence (2026-07-22): after restoring the stored OpenCode model through this fix, the actual browser history-resume turn completed with CHATMUX_E2E_OPENCODE_T2 value=OPENCODE_T1.