avoid buffer copy in generate_coding_shreds

[jbiseda]

Problem

Shredder::generate_coding_shreds allocates and zeros buffers for parity shreds, then copies parity data into shred buffers.

Summary of Changes

Use shred buffers directly.

Fixes #

[behzadnouri]

The tradeoff here is code simplicity and smaller api surface.
So I think it would be worth to see if these buffer copies do actually have a significant impact on validator runtimes.
Benchmarks rarely capture validator runtime and actual bottlenecks, so might be good to instead run a test gce cluster with and without this code and see if there is a significant gain; something like: #27510 (comment)

If there is no significant gain, I personally lean to have smaller api surface and keep the code simpler and more expressive to avoid introducing bugs.