Sending large body chunks (2GB+) are wrongly rejected

[seanmonstar]

Sending of chunks wrongly compares their size to the MAX_WINDOW_SIZE, which is just plain wrong.

[seanmonstar]

After some further investigation, I see why this is now in place. The "buffered data" amount is stored as a WindowSize, which represents the type from the HTTP2 spec, that has a max of u32::MAX >> 1. We could change things around to allow buffering more than that, but that's at least the why.