Use int normally, use int64 if needed

[Tougee]

A new try to optimize calculations, normally use int, use Int64 while calling load3(), load4(), FeToBytes(), FeCombine(), FeMul(), feSquare(), ScMulAdd(), there're some Int32 usages. Benchmark results are not fast enough, I'll make some detailed improvements later for object allocations.

[D00mch]

Why don't you merge it?

[Tougee]

Why don't you merge it?

Benchmark results not good.