Route default Foundation Models through Conduit

[christopherkarani]

Summary

• route the default Foundation Models provider path through the released Conduit runtime instead of the old private LanguageModelSession wrapper
• stop advertising streaming tool-call capability for the Foundation Models-backed Conduit path
• fix the MCP server service tests for the current MCP SDK text payload shape

Verification

• swift test --filter 'AgentDefaultInferenceProviderTests|ConduitProviderSelectionTests|InferenceProviderCertificationTests'\n- swift test

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6ae1df632f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Disable tool-call streaming path, not just capability flag

Setting supportsStreamingToolCalls: false here only changes reported capabilities; it does not prevent the provider from being treated as a streaming tool-call provider. ConduitInferenceProvider still conforms to ToolCallStreaming*, and Agent.executeToolCallingLoop decides useToolStreaming from protocol conformance (see Agent.swift around structuredToolStreamingProvider/promptToolStreamingProvider), so the default Foundation Models route will still enter generateWithToolsStreaming whenever streaming+observer are enabled. That means the new Foundation Models default can still take the streaming path you intended to disable, which can drop tool execution if streaming tool-call completion events are unavailable.

Useful? React with 👍 / 👎.