fix: keep Copilot Claude prompts user-final across continuation turns

[sryantich]

Issue for this PR

Closes #12337

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Issue #12337 reports GitHub Copilot Claude / Opus 4.6 failing with "This model does not support assistant message prefill. The conversation must end with a user message."

In opencode, that can happen when the autonomous loop continues after an assistant response or tool phase and the prompt sent to Copilot Claude still ends on a non-user turn. The max-step path could also reintroduce the same shape by appending an assistant-final message.

This change fixes that in two places. The session prompt loop now keeps Anthropic-style and Copilot Claude prompts user-final when it needs to continue, including the max-step edge case. ProviderTransform also adds a narrow last-line repair for providers that require user-final prompts, so Copilot Claude does not receive an assistant-final or tool-final transcript even if one slips through earlier prompt construction. The tests cover the prompt helper, the provider transform, and the Copilot message serializer.

How did you verify your code works?

• cd packages/opencode && bun test --timeout 30000
• cd packages/opencode && bun run typecheck

Screenshots / recordings

Not a UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍