Implement Baille-PSW instead of using F&J's hashtable

[JASory]

Forisek and Jancina's 16384-element hashtable requires 3 total SPRP tests, but modified variants of the BPSW can run in the same time as 2.5 fermat tests. So it is both faster and considerably more space efficient. Only when you call the 262144-element table to use 2 SPRP tests do you gain a speed improvement over the (modified) BPSW.

A reference implementation of a such a modified BPSW for 64-integers can be found in the Machine-prime source code. Directly translating it might be difficult particularly in the Montgomery arithmetic, but just replacing all the Montgomery forms with standard integer representations and operations will work.