fix: populate missing tee_id, tee_endpoint, tee_payment_address in StreamChunk.from_sse_data

[amathxbt]

Bug

StreamChunk declares three TEE metadata fields for identifying the enclave that served a streaming request:

tee_id: Optional[str] = None
tee_endpoint: Optional[str] = None  
tee_payment_address: Optional[str] = None

However, StreamChunk.from_sse_data() never reads these from the SSE data dictionary:

# BEFORE (broken) — tee_id, tee_endpoint, tee_payment_address silently dropped
return cls(choices=choices, model=..., usage=usage, is_final=is_final)

This means all three fields are always None in streaming responses, even when the TEE server sends them — making it impossible to:

• Identify which TEE served the request
• Verify the registered endpoint on-chain
• Look up the payment address

Fix

Expand the return cls(...) call to read the three missing fields from the SSE data dictionary:

# AFTER (fixed)
return cls(
    choices=choices,
    model=data.get("model", "unknown"),
    usage=usage,
    is_final=is_final,
    tee_signature=data.get("tee_signature"),
    tee_timestamp=data.get("tee_timestamp"),
    tee_id=data.get("tee_id"),              # was missing
    tee_endpoint=data.get("tee_endpoint"),  # was missing
    tee_payment_address=data.get("tee_payment_address"),  # was missing
)

Impact

• Streaming TEE metadata is now accessible — callers can identify and verify the serving TEE from the final chunk
• Non-streaming path (TextGenerationOutput) already populates these fields correctly — this aligns the two code paths
• Backward compatible: fields still default to None if not present in the response

[amathxbt]

@adambalogh