feat: add non-buffered response streaming

[GrahamCampbell]

Allow streaming output operations to opt out of response buffering so large downloads can be consumed without temp storage while keeping buffered behavior as the default. My use case is streaming a large CSV file out of S3. Without this change, I have to go around the SDK and create a presigned URL and download it myself using another HTTP client. Without the ability to turn off buffering, the entire file gets written to /tmp.

[GrahamCampbell]

Hmmm, tough to test due to the cyclic nature of needing a newer version of core...

[jderusse]

I would avoid creating precedence by adding a new custom @responseBuffer parameter.

I didn't yet read the implementation, but some replacement suggestion in mind

1. Is there a reason for buffering the output? Couldn't we alway stream without buffer? (rewind come in my mind, but do we have something else? could we let end-user buffer the output if they want to rewind. And release a new Major version that don't buffer)
2. or buffer when the response is small, and that size is configurable with something like new S3Client(['doNotBufferWhenSizeExceed' => '3Gib'])
3. Or maybe more simple : can't we add a new response->getContentAsResource(buffer: false) to achieve the same?

[GrahamCampbell]

Yes, we do want to buffer the output in general by default (and by default, small responses buffer entirely in memory without spilling to disk anyway). Without buffering it, we can't read it more than once, and there are many situations where we would want to do that.