Optimize Extend trait to have BytesMut::extend(&mut self, &[u8) compile to a memcpy when possible #848

I am currently working on a library that copies a data slice received from a QUIC library into a passed in buffer. I didn't want to force a specific buffer type so I have the passed in buffer implement for<'a> Extend<&'a u8>. Currently if a Vec is passed in as the buffer it compiles down to a memcpy. This makes sense because Vec's docs state:

This implementation is specialized for slice iterators, where it uses copy_from_slice to append the entire slice at once.

Meanwhile BytesMut copies data byte-by-byte, even when compiled in release mode.

Here's a playground link with the following source code to verify that BytesMut copies data byte-by-byte:

use bytes::{Bytes, BytesMut};

#[inline(always)]
fn read_into_buf(bytes_from_network: &[u8], buf: &mut impl for<'a> Extend<&'a u8>) -> () {
    buf.extend(bytes_from_network);
}

#[unsafe(no_mangle)]
fn with_vec(bytes_from_network: &[u8]) {
    let mut into = Vec::new();
    read_into_buf(bytes_from_network, &mut into);
}

#[unsafe(no_mangle)]
fn with_bytes_mut(bytes_from_network: &[u8]) {
    let mut into = BytesMut::new();
    read_into_buf(bytes_from_network, &mut into);
}

The BytesMut Extend implementation currently has this as a TODO comment:

I was wondering if there was a reason this optimization has not been implemented yet. It would allow extend calls, like the one in my library, to turn into memcpy instead of copying data byte-by-byte.