head: port splice fast-path for -c as seen as wc -c

[oech3]

I noticed that we have

coreutils/src/uu/wc/src/count_fast.rs

Lines 40 to 77 in 572f3a2

	/// This is a Linux-specific function to count the number of bytes using the
	/// `splice` system call, which is faster than using `read`.
	///
	/// On error it returns the number of bytes it did manage to read, since the
	/// caller will fall back to a simpler method.
	#[inline]
	#[cfg(any(target_os = "linux", target_os = "android"))]
	fn count_bytes_using_splice(fd: &impl AsFd) -> Result<usize, usize> {
	let null_file = OpenOptions::new()
	.write(true)
	.open("/dev/null")
	.map_err(|_| 0_usize)?;
	let null_rdev = stat::fstat(null_file.as_fd()).map_err(|_| 0_usize)?.st_rdev as libc::dev_t;
	if (libc::major(null_rdev), libc::minor(null_rdev)) != (1, 3) {
	// This is not a proper /dev/null, writing to it is probably bad
	// Bit of an edge case, but it has been known to happen
	return Err(0);
	}
	let (pipe_rd, pipe_wr) = pipe().map_err(|_| 0_usize)?;
	let mut byte_count = 0;
	loop {
	match splice(fd, &pipe_wr, SPLICE_SIZE) {
	Ok(0) => break,
	Ok(res) => {
	byte_count += res;
	// Silent the warning as we want to the error message
	#[allow(clippy::question_mark)]
	if splice_exact(&pipe_rd, &null_file, res).is_err() {
	return Err(byte_count);
	}
	}
	Err(_) => return Err(byte_count),
	}
	}
	Ok(byte_count)
	}

I think head -c can splice too.