Pydantic AI adds a GitHub Copilot provider

Pydantic AI 2.42.0 adds GitHubCopilotProvider, letting agents call GitHub Copilot's OpenAI-compatible endpoint using a Copilot subscription instead of a separate API key. For anyone already paying for Copilot, that's one less credential to manage.

The release also changes how DeferredToolResults.approvals behaves. It now rejects invalid values instead of accepting them, per Pydantic AI's 2.42.0 release notes. Code that was passing malformed approval payloads and having them silently accepted will start raising errors on upgrade.

The stricter validation changes behavior for existing users. The provider doesn't, since nobody has to touch it to keep their agent running.

If your agent setup builds approval payloads dynamically, whatever was wrong before and never surfaced is about to throw. Run your test suite against 2.42.0 before touching production, specifically around tool-approval flows.

The Copilot provider itself is straightforward. Point Pydantic AI at the Copilot endpoint, skip the extra API key. It's a convenience feature for teams already inside GitHub's ecosystem, not a new capability class.