fix: add missing __aiter__ to TextGenerationStream for async for loops

[amathxbt]

Bug

TextGenerationStream implements __anext__ for async iteration but is missing the required __aiter__ protocol method.

Any code using async for chunk in stream: when _is_async=True will crash with:

TypeError: 'TextGenerationStream' object is not an async iterable

This silently breaks all async streaming usage — a critical runtime regression.

Root Cause

Python's async iteration protocol requires both __aiter__ (to return the iterator) and __anext__ (to yield items). Only __anext__ was implemented.

From the Python data model docs:

An object must implement both __aiter__ and __anext__ to be an async iterable.

Fix

Added __aiter__ method that validates the _is_async flag and returns self:

def __aiter__(self):
    """Return async iterator (required for async for loops)."""
    if not self._is_async:
        raise TypeError("Use __iter__ for sync iterators")
    return self

Impact

• Fixes async for chunk in stream: pattern documented in SDK examples and README
• No behaviour change for sync iteration
• Low risk, zero-dependency change

Testing

async def test_async_stream():
    stream = TextGenerationStream(_iterator=some_async_gen(), _is_async=True)
    async for chunk in stream:  # Would previously raise TypeError
        print(chunk)

[amathxbt]

Hey @adambalogh this is something critical you need to take action quickly guys !

[adambalogh]

Thanks for your contribution!

[amathxbt]

There's still some pr opened can you check them on your X dm or let me provide them here too

Hey team,Just pushed 4 bug fix PRs to the OpenGradient-SDK repo — all critical issues found during a code audit. Here's a quick summary:🔴 PR #196 — Async Streaming Completely Broken
TextGenerationStream was missing the aiter method, meaning async for chunk in stream: would crash with a TypeError for anyone using async streaming. Fixed ✅
🔗 https://t.co/0ZwOlWzmNo

🔴 PR #197 — TEE Metadata Lost in Streaming Responses
StreamChunk.from_sse_data() was silently dropping tee_id, tee_endpoint, and tee_payment_address — always returning None even when the server sent them. This made it impossible to verify which TEE served a streaming request. Fixed ✅
🔗 https://t.co/LcZaGCpMFI

🟠 PR #198 — run_with_retry Silently Returns None
Passing max_retries=0 caused an empty loop and a silent None return instead of executing the transaction — leading to confusing downstream crashes. Added proper validation + safety net. Fixed ✅
🔗 https://t.co/Ymdk9FAjnr

🟡 PR #199 — Public Types Missing from all
InferenceResult, ModelOutput, ModelRepository, and FileUploadResult were imported but not listed in all — breaking wildcard imports and IDE auto-complete. Fixed ✅
🔗 https://t.co/2RraRDHDpx Would appreciate a quick review when you get a chance 🙏