better decision of how many blocks to request from a peer

[ckerr]

In main right now it appears to request too many blocks and then periodically cancel the excess requests. In theory this is not terrible but in practice it seems to cause speed oscillations -- cancelling fewer requests is giving me faster, steadier speeds. libtransmission should make more of an effort to calculate what the 'ideal' number of pending requests is so that on average we need to cancel fewer of them.

[LaserEyess]

Some prior art: https://github.com/mandreyel/cratetorrent/blob/master/DESIGN.md#download-pipeline

[ckerr]

https://github.com/transmission/transmission/tree/refactor/block-req-allocation has the work so far.