Fuzzing campaign fixes

[mratsim]

This PR

• Adds test cases for squaring bugs
  • Fix Modular Square Root (spurious?) failure for FKM12_447 #30, Fused modular square root on 32-bit - wrong "isSquare" result #42, Fused modular square root on 32-bit - wrong "isSquare" result #43, Randomized test failure EC add G2 #60, Randomized test failure sqrt mersenne127 #61, Randomized test failure negate in sqrt Mersenne 127 #62, Randomized test failure EC Add G2 associative #64, Randomized testing failure: EC Add G2  #65, Randomized test failure - ARM - G1 Infinity #67, (Randomized testing failure: EC Add G2  #65 has 3 cases.)
• "Fix" them by using regular multiplication instead.

Reactivating the squaring which would reenable a significant 20% performance improvement is tracked in #68.
Note that the error is an off-by-one in the least significant bit in the Montgomery representation.
It happens in both the generic squaring algorithm and the fast squaring from Consensys.

I suspect a carry is missing from the literature, though the literature on Montgomery squaring is pretty scarce.