remove the generic "nbit" parameter from the number scalar types

[jorenham]

Since #136, the generic "nbit" parameter has no use anymore for the these abstract scalar types. This type parameter is only still used in _numtype to shorten type-alias definitions.

From a type-theoretical perspective, this type parameter is invalid: It does not describe anything (yet is invariant), and should therefore be removed.

Since NumPy 2.2 it is no longer a required type parameter, so that should make the transition less painful.