Optimize `VarInt`/`VarLong` encode and decode procedures

[rj00a]

Varints are prolific in the Minecraft protocol. It makes sense to try and optimize them as best as we can. The current implementation is straightforward but there is room for improvement.

See:

• how fast can you encode a varint?
• Moulberry's decode and fast branchless encode functions

Benchmarks must be made with criterion to establish confidence.

• [x] Optimize encode functions (#227)
• [ ] Optimize decode functions

[rileysu]

Would it increase performance just to unwrap the loop in this case? Other than that I am not sure how to convert the Read to a statically sized buffer that can deliver some of these performance improvements.

[rj00a]

We don't need to use Read. &mut &[u8] should be used instead.