Missing support for adaptive thinking for Anthropic models

[JakubK44]

Background

1. Per Anthropic docs, adaptive thinking "is the recommended way to use extended thinking with Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6".
2. Also per Anthropic docs, effort should be used instead of thinking token budget for Opus/Sonnet 4.6+.
3. Currently, the SDK's SessionConfig.reasoningEffort accepts the following values: export type ReasoningEffort = "low" | "medium" | "high" | "xhigh";. They correspond to a body.thinking object containing { type: "enabled", budget_tokens: XYZ } where XYZ is 1024, 2048, 4096, 8191 (not 8192!) for the respective reasoning efforts.

Issue

SDK still uses thinking.budget_tokens for models where it should be using adaptive thinking and output_config.effort (Opus/Sonnet 4.6+)

Repro version

"@github/copilot-sdk": "0.2.2"
"@github/copilot: "1.0.26"