[FEATURE]: Forward promptCacheKey to GitHub Copilot GPT (Responses) requests #61936

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

On v2.2.5, Copilot-hosted GPT models never receive a prompt_cache_key. That leaves them without the session affinity anchor that OpenAI-native models already get.

• OpenAI-native already gets it. session/model-request.ts sets request.promptCacheKey from session affinity, and packages/ai/src/protocols/openai-decisions.ts forwards it.
• Copilot GPT doesn't. Those models (gpt-* and other non-Claude models on github-copilot) skip packages/ai. github-copilot/models.ts routes them to Provider.aisdk("@ai-sdk/github-copilot"), and packages/core/src/aisdk.ts never reads request.promptCacheKey.
• The Responses converter would honour it. github-copilot/responses/openai-responses-language-model.ts already reads providerOptions.copilot.promptCacheKey; nothing on the AI-SDK path fills it in.
• The compatibility flag can't help. Setting compatibility.supportsPromptCacheKey for github-copilot has no effect, because only packages/ai protocol code reads it.

The change I'm asking for is to merge request.promptCacheKey into providerOptions.copilot for @ai-sdk/github-copilot Responses calls, gated on supportsPromptCacheKey. This mirrors the OpenAI path. It could live in aisdk.ts or in the github-copilot plugin's language hook.

Benefit: long agent sessions on Copilot GPT keep landing on the same cache shard, the same way OpenAI-native sessions do. #59868 shows Copilot GPT already gets implicit cache reads; this keeps them stable across turns.

I have not verified that Copilot's /responses endpoint accepts prompt_cache_key. Related: #59868, #55244, #51931.