encode compute_size method is very slow.

[rukai]

In my usage of kafka-protocol I just directly use Encodable::encode without first reserving any bytes in the BytesMut.
Given that there is an Encodable::compute_size method I figured it must be faster to first use that first and then reserve the needed space.
However doing that resulted in my benchmarks regressing!

I guess the compute_size method is just more expensive than avoiding reallocating.
I'll do a further investigation myself, as I really would like compute_size to improve performance, currently this issue is mostly a headsup to others that the problem exists.

[rukai]

I fixed a bug in my code and now I'm seeing neutral results or possibly a slight increase in performance 😅
Sorry for the noise!