fix(runtime): filter broken models from OpenCode Free probe candidates

[SummerC0zyR0ck]

Summary

Fixes #3449

OpenCode Free connection tests could still probe a quarantined model saved in enabledModelIds or defaultModel, even though the model catalog and execution authorization already rejected the same brokenModelIds entry.

This change makes the connection-test candidate path use the same quarantine authority as the catalog and execution paths:

• filter quarantined enabled/default models before probing;
• preserve the existing fallback order for healthy OpenCode Free models;
• return a clear No model to test result when every candidate is quarantined;
• keep explicit model tests unchanged;
• do not introduce automatic in-turn failover, silent model switching, or new
  provenance behavior.

Verification

Focused Runtime regression coverage passed: 10 tests, 0 failures. The affected OpenCode Free connection-test suite verifies:

• quarantined models persisted in enabledModelIds are excluded from probing;
• a quarantined persisted defaultModel does not block selection of a healthy fallback;
• when all enabled and fallback models are quarantined, no network request is issued and the result is No model to test;
• customized OpenCode Free connections preserve their existing candidate order;
• a successful fallback is selected after an error envelope or malformed completion from the first candidate;
• explicitly requested models are tested once and do not trigger fallback probing;
• anonymous OpenCode Free requests omit the Authorization header;
• reasoning completions with content: null remain valid;
• all fallback attempts share one overall timeout budget.

Build and static checks also passed:

• Storage build;
• Runtime build;
• Runtime typecheck;
• Biome checks for the changed Runtime source and test files.

The full repository test suite was not run.

AI use

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: OpenAI Codex investigated the defect and fixed, authored the implementation review and Runtime/UI/TUI regression tests.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — connection tests no longer probe models quarantined by the provider
  registry; explicit model testing and manual model switching are unchanged.
• No