fix(middleware-flexible-checksums): buffer stream chunks to minimum required size

[kuhe]

helps with #6859

Attempts to buffer input stream chunks to a minimum required size as required by S3, as described in https://docs.aws.amazon.com/AmazonS3/latest/API/sigv4-streaming.html.

This is for input streams e.g. S3 PUT Object or Part.

With checksumming on, user streams go through a filter called AWS chunked encoding. This process annotates every stream chunk with its size and thus the chunks are sent to S3 as-is in whatever size the user provides them, instead of Node.js being able to automatically buffer smaller chunks.

S3 has always required a chunk size of 8kb, preferably 64kb, but due to automatic buffering this was rarely an issue prior to the introduction of default checksums.

As a mitigation / convenience feature, the JS SDK can automatically buffer streams for the user before undergoing chunked encoding. This incurs a small performance penalty that can be avoided by the user if they provide streams with larger chunks.

• needs codegen update from feat(util-stream): buffering of stream chunks smithy-lang/smithy-typescript#1523

[kuhe]

• unit
• integ
• e2e
• cucumber (s3 tag especially)