Simplify <Vec<u8> as BufMut>::bytes_mut()

[danburkert]

Removes a call to UninitMut indexing, which can panic, from Vec::<u8>::bytes_mut(). I came across this while looking at a performance issue encountered after moving to bytes 0.6. This change makes a microbenchmark in prost slightly faster, although there's still a big delta between 0.5 and 0.6. See danburkert/prost#381.

[danburkert]

Added a second commit which adds an #[inline] for Vec::put_slice, which closed most of the remaining perf difference between 0.5 and 0.6. Will have another PR to add inlines to UninitSlice to close out the rest.

[taiki-e]

@danburkert Could you resolve merge conflicts?