Single perfect radix encoding specialization

[oscbyspro]

From oscbyspro/Numberick#31

Only one specialized encoding method is required for "imperfect" radices, wheras one specialized encoding method per integer type is required for "perfect" radices because the "perfect" radix chunk sequence is a Slice<some NBKDoubleWidth>. I may reduce this to a single specialization by using the buffer pointer from withContiguousStorage(_:) instead. This approach also turns out to be about 10% faster. Win-win.

Note: with big-endian reverse() it is still faster than the current approach.