Get raw finish reason from provider

[egoist]

Feature Description

when using google gemini, ai sdk would treat IMAGE_SAFETY and some other reasons as unknown finish reason in streamText

data: {"candidates": [{"finishReason": "IMAGE_SAFETY","index": 0}],"usageMetadata": {"promptTokenCount": 8,"totalTokenCount": 8,"promptTokensDetails": [{"modality": "TEXT","tokenCount": 8}]},"modelVersion": "gemini-2.0-flash-exp"}

Use Cases

need the underlying finish reason to display warnings to user properly

Additional context

No response

[lgrammel]

#5311 maps IMAGE_SAFETY to content-filter. Leaving the issue open since the raw finish reason is not exposed.

[lgrammel]

Release @ai-sdk/[email protected]