BytesParsing - change handling of calldata slicing and checkLength

[nonergodic]

Two changes:

First, the various calldata versions of slice now return bytes calldata instead of bytes memory.

This rectifies a previous mistake of mine where I didn't realize that I could use assembly to implement an unchecked version of slice that does what I want. I figured that creating a bytes memory is efficient enough since it can use CALLDATACOPY but failed to consider that pushing things from the calldata domain to the memory domain makes for ugly and inefficient code if subsequent parsing is necessary.

Second, checkLength now takes the encoded length as the first argument instead of the encoded bytes directly. The rationale is that it's way too easy to write someCalldataBytes.checkLength(expectedLength); instead of using the appropriate checkLengthCd version, which would quietly allocate memory and copy the entire someCalldataBytes just to do the length check. Additionally, it's also more consistent with how checkBound works.

It is now recommended to also have a using {BytesParsing.checkLength} for uint; line at the beginning of the contract (along with using BytesParsing for bytes;) and check via encoded.length.checkLength(expectedLength); independent of whether encoded has memory or calldata location.