Some operations on `xsd:Integer` are unecessarily expensive

[liss-h]

For example: casting an xsd:Integer to string requires the following in most cases

1. un-inline the integer value into a int64_t (this is true for most integers)
2. create a boost::multiprecision integer out of it
3. call the to string function for the boost::multiprecision integer

In theory it would be more optimal if we did not construct the boost integer, instead just converting the uninlined int64_t to string.

Not sure on the implementation yet.

[liss-h]

This could also be solved by switching away from boost::multiprecision and using a library with less overhead for small integers