Don't generate redundant assertion in doPtrAddOffset when offset = 0.

[brianhuffman]

Now doPtrAddOffset only asserts pointer validity of the result when
it actually produces a new pointer; if the offset is a concrete zero,
then it returns the original pointer which was already assumed to be
valid.

Fixes GaloisInc/saw-script#585.

[brianhuffman]

I'm actually not entirely convinced that this is the best way to solve the problem. It might also make sense to detect getelementptr array indexing at index 0, and just avoid generating the LLVM_PtrAddOffset instruction in the Crucible CFG in that case.

[brianhuffman]

After talking with @robdockins, I think we should probably solve this problem a different way. This PR makes doPtrAddOffset have a kind of weird implicit precondition that either the offset is nonzero OR the input pointer was already asserted to be valid.

It will be better to change the llvm-to-crucible translation to skip the LLVM_PtrAddOffset when used with size 0. I'll make a separate pull request following that approach.