The main benefit is reduced memory allocations, as we now count the tokens during the validation step and then work off of a flat array without nesting children.

I have also slightly loosened the coupling between tokenizer and deserializer.

Lastly there has been some housekeeping with naming and file locations.

Saves some time and allocation.

…ork with XUnit.

- Do not get the PhpDataType and switch on that, intead directly switch on the relevant input byte.
- In the validator: Increment the token count only in the array and object methods.
- Tokenizer: Add AggressiveOptimization hint to GetToken().

Also: For small inputs, we can stackalloc the byte array.

Updated changelog to reflect the performance status.

This reduces memory allocation for doubles, bools and floats, as they can be directly evaluated from the raw bytestream.  That also speeds up deserialization by a lot.

We also no longer have to give the Tokenizer the input encoding.

There are probably some further optimizations to be had in the deserializer -> DeserializeTokenFromSimpleType() method too.