fix(openai): use empty instructions fallback

[dahetaoa]

Summary

• stop injecting the semantic default You are a helpful coding assistant. when Responses instructions is missing
• normalize missing, nil, blank, or non-string instructions to an empty string instead
• keep existing non-empty instructions and extracted system messages unchanged

Why

Direct testing against the ChatGPT Codex internal endpoint showed that omitting instructions returns 400 {"detail":"Instructions are required"}, while sending "instructions": "" succeeds. This keeps the required field without adding an unintended system prompt.

Tests

• go test -race ./internal/service -run 'TestApplyCodexOAuthTransform|TestOpenAIGatewayService_OAuthResponsesUsesEmptyInstructionsWhenMissing|TestOpenAIGatewayService_OAuthMessagesBridgeDoesNotInjectDefaultInstructions|TestOpenAIImageGeneration'
• go test ./internal/service
• go test ./internal/pkg/openai ./internal/pkg/apicompat

[github-actions]

All contributors have signed the CLA. ✅
_{Posted by the CLA Assistant Lite bot.}

[dahetaoa]

I have read the CLA Document and I hereby sign the CLA