Fix missing signed empty end frame on event stream close

[alexgromero]

Issue #, if available:
awslabs/aws-sdk-python#46

Description of changes:
Fix AWSEventPublisher.close() to send a signed empty end frame to signal stream completion.

SigV4 signed event streams require a final signed empty message before the HTTP body stream closes. This fix:

1. Adds sign_empty() method to the EventSigner protocol in smithy-core.
2. Updates AWSEventPublisher.close() to call sign_empty() when sending the termination frame in smithy-aws-event-stream.

This eliminates the following errors:

• Transcribe Streaming: BadRequestException: A complete signal was sent without the preceding empty frame.
• Bedrock Runtime: ValidationException: Invalid input request, please fix your input and try again.

Related PR: #692

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[Alan4506]

Just curious: with this change applied, are we able to modify some existing integration tests to remove await asyncio.sleep statements? For example, Bedrock Runtime and Transcribe Streaming? When I wrote tests for new clients (e.g., Lex Runtime V2 and Q business), I also had similar concerns - tests would fail without await asyncio.sleep statements.

[Alan4506]

Approved. However, there are some other issues when I test this with the Transcribe Streaming integration tests and the simple_file.py example. The end frame fix resolves the "complete signal was sent without the preceding empty frame" error, but the asyncio.sleep calls before close() seem still needed. Without them, it returns aws_sdk_transcribe_streaming.models.BadRequestException: Could not decode the audio stream that you provided. Check that the audio stream is valid and try your request again. I wonder if this is expected behavior or something that needs a separate fix.

[jonathan343]

Could we wrap the signed-empty-frame write in a try/finally so the writer is always closed?

It may also be worth making the test assert that the writer is closed, or using AsyncBytesProvider to decode the written frame and verify the terminator payload is actually empty.

[alexgromero]

Addressed in d1fc057.

[jonathan343]

Do you know if this is safe for even non-bidirectional APIs?

[jonathan343]

I just realized all AWS operations that use input streaming are also bi-directional

[alexgromero]

Yep, thanks for confirming! Also worth noting the Java SDK sends the trailing empty frame unconditionally for all signed event streams, no bidi check, so this appears to be part of the SigV4 protocol itself rather than a bidi-specific behavior.