Fix Bin/Hex parsing of BigInteger for powers of 2

[kzrnm]

dotnet#95543 (comment)

I fixed Bin/Hex parsing of BigInteger for powers of 2.

// -1 << 32
BigInteger.Parse("100000000000000000000000000000000", NumberStyles.BinaryNumber);
BigInteger.Parse("F00000000", NumberStyles.HexNumber);

// -1 << 64
BigInteger.Parse("10000000000000000000000000000000000000000000000000000000000000000", NumberStyles.BinaryNumber);
BigInteger.Parse("F0000000000000000", NumberStyles.HexNumber);

[kzrnm]

I removed the (int)leading == int.MinValue condition and changed it to call new BigInteger(int).
It avoids allocation by using BigInteger.s_bnMinInt.

[huoyaoyuan]

Interesting, I didn't realize that BigInteger(int) constructor has an optimization to avoid the allocation for int.MinValue.
In previous commit I delegated to that constructor to handle MinValue. I'll update the comment.

[huoyaoyuan]

Sorry, I didn't mention this was in my fork.

[huoyaoyuan]

Thank you for pointing out my mistakes!

I'm going to think more about optimization and measure the performance of each case.

[huoyaoyuan]

Interesting, I didn't realize that BigInteger(int) constructor has an optimization to avoid the allocation for int.MinValue.
In previous commit I delegated to that constructor to handle MinValue. I'll update the comment.

[huoyaoyuan]

I missed that the bits of negative result needs negation here.

The non-zero case for negative and positive can be combined in bit hack: leading ^ signBits - signBits. The zero case needs [0, 1].

[huoyaoyuan]

A new digit is required only if all the trailing digits are zero.
It's really a headache for allocation management.