further optimize BytesParsing assembly and gas cost

[nonergodic]

what it says on the tin:
instead of using the uint function version to read bytes and then have casts that also have to shift stuff around, it's more efficient to just read the bytes directly and have Solidity take care of cleaning up the dirty bytes when returning bytes<n>

More explicitly, when compiling e.g. this sample contract:

contract ByteCode {
  function foo() public payable returns (bytes16 ret) { unchecked {
    bytes memory buf = abi.encodePacked(
      bytes32(0x3300000000000000000000000000000000000000000000000000000000000044)
    );
    (ret, ) = BytesParsing.asBytes16Unchecked(buf, 3);
  }}
}

The original assembly is:

  /* \"src/BytesParsing.sol\":19598:19711  assembly (\"memory-safe\") {... */
0x13
add
mload
  /* \"src/ByteCode.sol\":95:394  contract ByteCode {... */
0x80
shl
not(0xffffffffffffffffffffffffffffffff)
and
dup2
mstore

While the new, optimized version gives:

  /* \"src/BytesParsing.sol\":20679:20803  assembly (\"memory-safe\") {... */
0x23
add
mload
not(0xffffffffffffffffffffffffffffffff)
  /* \"src/ByteCode.sol\":95:394  contract ByteCode {... */
and
dup2
mstore