Pool allocate arrays instead of slices

[H1JK]

This is inspired by https://go-review.googlesource.com/c/net/+/539915

This remove the allocation for the slice header in sync.Pool.New and putDataBufferChunk.
It also divide the number of allocations kept alive by the pool.

Because Go interfaces are optimized with pointer values, this would also remove 1 alloc in Get() and make buffer put-and-get process zero-alloc.

The binary size may increase by about 7KB as a trade-off, but the execution speed is slightly improved.

[nekohasekai]

Merged in ed23253 and a5c920a