Use more decimals of precision, internally

[PaulRBerg]

We need to take the inverse of the input number x in the following two cases:

1. log2
2. exp2

That is, when x is negative. The problem is that the inverse limits the precision of the input to maximum 36 decimals, otherwise the result would not fit in the SD59x18/ UD60x18 fixed-point representation.

One way to make this better is to increase the precision that is used internally within PRBMath, similarly to how DSMath uses RAY internally (which has 27 decimals of precision) and WAD externally (which has 18 decimals of precision).