Introduce a reusable RequestBody and ensure Body::__toString() rewinds properly

[opengeek]

This introduces a reusable RequestBody implementation that by default copies the php://input stream, which can only be read one time, to a php://temp stream that can be re-wound and re-read.

Also makes sure Body::__toString() does not throw any exceptions.

This should address issue #1386.

[JoeBengalen]

Will stream_copy_to_stream be faster than appending each string being read to a cache property like phly does?
https://github.com/phly/http/blob/master/src/PhpInputStream.php

[opengeek]

I tried both approaches and did some profiling with them but could not find any significant difference with small, basic payloads. The approach of keeping it an actual stream rather than caching the content should provide more flexibility within middleware usage however, with only slightly more memory usage. Copying it to the temp stream and discarding the original input stream should theoretically save memory usage however, with larger payloads since it uses temp files after 5Mb of memory is used.

I may try to find a way to test this with some larger payloads to better understand the impact. And now I wonder if I should call this implementation ServerRequestBody.

[akrabat]

I don't think we need to accept a $stream parameter here. This is a specialist class solely for handling php://input

[akrabat]

Can we simplify RequestBody to something like:

class RequestBody extends Body
{
    /**
     * Create a new RequestBody.
     */
    public function __construct()
    {
        $stream = fopen('php://temp', 'w+');
        stream_copy_to_stream(fopen('php://input', 'r'), $stream);
        rewind($stream);

        parent::__construct($stream);
    }
}

i.e. what do we gain with your current implementation over this?