[WIP] Skewed RNGs that trigger corner cases

[mratsim]

This PR addresses #53.

• An uniform RNG has a hard time triggering carry paths, especially with 64-bit word, the probability is 2^-64 * 2^-64 = 2^-128.

• A significant amount of algebra bugs in established libraries like OpenSSL are due to carries.

• Most fuzzing tools like AFL or LibFuzzer are branch coverage based, but constant-time code is branchless so it's likely that they won't focus on the carry path as it's indistinguishable from the normal path from a Control Flow Graph perspective.

• Add a RNG biaised towards high Hamming Weight words

• Add the same RNG as libsecp256k1 that produces long sequences of 1111... and 0000...
  and found CVE-2014-3570 in OpenSSL and would likely have found similar bug in TweetNaCK: https://gist.github.com/CodesInChaos/8374632

• Use them in testing, in particular check if they can trigger the same edge cases as Modular Square Root (spurious?) failure for FKM12_447 #30, Fused modular square root on 32-bit - wrong "isSquare" result #42 and Fused modular square root on 32-bit - wrong "isSquare" result #43 that are targeted for solving in Fuzzing campaign fixes #58

[mratsim]

New test failures reported in #60 #61. #61 in particular is probably reproducing the #42 #43 sqrt bug on only 2 limbs which is nice confirmation of the power of this approach.

[mratsim]

Test failure raised: #62