do 2 MR rounds for numbers >=2048bits

[sjaeckel]

While looking at the coverage results in #268 I think I found an inconsistency... nothing bad but it still stood out while looking at the code...

mp_prime_rabin_miller_trials:38 leads to 2 MR trials for numbers > 4096 bits

@czurnieden is that change fine that not only 4096bit sized numbers only have to do 1 MR but also numbers >4096bits?

Edit: changed to do always at least 2 Rounds

[czurnieden]

is that change fine that not only 4096bit sized numbers only have to do 1 MR but also numbers >4096bits?

Yes.

But now that I think about it: that is the theoretical value. Carefully calculated but still only a theoretical value.
Could we set the minimum to (2) two to sooth my conscience?

[czurnieden]

Thanks!