Fix TokenRouter Claude OpenCode protocol

[ShuxinLin]

Description

Configure OpenCode to use the Anthropic AI SDK protocol for TokenRouter Claude models instead of routing those requests through the OpenAI-compatible provider.

Fix Details

• Detect tokenrouter/anthropic/... model IDs in the OpenCode runner provider config.
• Use @ai-sdk/anthropic for those Claude models while keeping other router models on @ai-sdk/openai-compatible.
• Add unit coverage for tokenrouter/anthropic/claude-opus-4.8.
• Update CLI help to describe the TokenRouter Claude protocol behavior.

Impact on Benchmarking

• No change to baselines: This fix only improves stability/performance.
• Baseline change: This fix corrects a scoring error. (Please provide "Before vs. After" results).

Related Issues

• Fixes: OpenCode Claude runs fail through TokenRouter and direct Anthropic model ID #426

Verification Steps

1. Run the following command: uv run pytest src/agent/opencode_agent/tests/test_runner.py -v
2. Focused OpenCode runner tests passed: 21 passed.
3. Run Ruff on touched files: uvx ruff check ... and uvx ruff format --check ...
4. Ruff passed: all checks passed, 3 files already formatted.
5. uv run pytest src/ -v -k "not integration" was also run from a clean worktree, but failed on unrelated existing issues outside this change path: direct LLM JSON spacing, missing google.protobuf for observability file exporter tests, and an IoT mock DB assertion.

Checklist

• I have added tests that prove my fix is effective.
• My code follows the project's Ruff formatting and linting rules.
• I have signed off my commits (DCO).