feat: add async streaming download support for daily extract files

[be-ant]

Related

Companion to #160 (sync streaming download), which documents the memory problem and POC evidence in detail.

Problem

The async client does not expose a get_daily_extract_file method at all (it is listed as a NotImplementedError stub per #151), and when it is eventually implemented it should not repeat the sync client's mistake of buffering the full response in memory.

httpx — the async HTTP library used by the async client — supports streaming via client.stream(...) as an async context manager. The async equivalent must use this pattern instead of await response.aread() or response.content.

Proposed Solution

Add get_daily_extract_file_stream(date, filename, chunk_size=8192) to the async client as an async method that:

1. Uses httpx.AsyncClient.stream("GET", url, ...) as an async context manager
2. Yields chunks via response.aiter_bytes(chunk_size) (async generator)
3. Allows the caller to control where bytes land (file, S3, GCS, etc.) without any intermediate buffering

Example usage (desired API)

async with ofsc.get_daily_extract_file_stream(date="2024-01-15", filename="extract.zip") as stream:
    async with aiofiles.open("extract.zip", "wb") as f:
        async for chunk in stream:
            await f.write(chunk)

Or for direct streaming to S3 via a multipart upload:

async with ofsc.get_daily_extract_file_stream(date="2024-01-15", filename="extract.zip") as stream:
    async for chunk in stream:
        await upload_part(chunk)

Impact

Daily extract ZIPs can be 50+ MB. In Lambda (128–512 MB) or any async microservice, buffering the full file wastes memory and risks OOM. The sync POC showed peak memory dropped from ~108 MB → ~55 KB when streaming (see #160 for details). The async path should achieve the same.

Implementation Notes

• httpx streaming requires the request to stay open for the duration of iteration; the method should be an async context manager (or return one) so connections are properly closed
• The wrap_return decorator used by sync methods likely does not support async streaming — this method may need to bypass it similarly to the sync counterpart in feat: add streaming download support for daily extract files #160
• Content-Length is present in OFS responses, enabling progress tracking if desired