• Add robust availability checks for Apple Intelligence (iOS 18 / macOS 15) before using system language model
• Implement health check via LanguageModelSession to validate model responsiveness (“ping” -> “pong”)
• Provide synchronous completion API appleFMComplete(prompt:) returning model response text
• Implement streaming completion API appleFMStream(prompt:) that yields incremental deltas
• Fix optional handling in streaming: unwrap partial.content.text safely and remove force unwraps
• Correct delta computation using unwrapped current string and maintain last state
• Add graceful fallback to single-shot completion if streaming fails
• Provide stubbed implementations when USE_FOUNDATION_MODELS is disabled to ensure build stability across configs
