read_u32 & unaligned reads

le.rs

/// Read a `u32` from a byte sequence, in litte-endian order
/// 
/// Consider usage with the `arrayref` crate.
pub fn read_u32(bytes: &[u8; 4]) -> u32 {
    // TODO: check; also isn't the only code difference the read type?
    // TODO: but since this is unused we could just remove it!
    // On platforms where unaligned reads are safe, do a simple read
    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] {
        unsafe{ *(bytes as *const [u8; 4] as *const u32) }.to_le()
    }
    #[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
    unsafe {
        let mut v: u32 = mem::uninitialized();
        ptr::copy_nonoverlapping(
            bytes.as_ptr(),
            &mut v as *mut u32 as *mut u8,
            mem::size_of::<u32>());
        v.to_le()
    }
}

/// Read a `u64` from a byte sequence, in litte-endian order
/// 
/// Consider usage with the `arrayref` crate.
pub fn read_u64(bytes: &[u8; 8]) -> u64 {
    // TODO: as read_u32
    unsafe{ *(bytes as *const [u8; 8] as *const u64) }.to_le()
}

@pitdicker what do you think — the original (as read_u64 here) is potentially unsafe?

Also, I checked dhardy/master and found only one use (MockAddRng), and it's possible to use the slice versions instead anyway, so shall we just remove these functions?