Teach Clang/LLVM that `rem_euclid` is always positive

Godbolt: https://godbolt.org/z/3d15c66s4
Alive2: https://alive2.llvm.org/ce/z/gpzmSn

Clang/LLVM is able to deduce that `rem_euclid(x, positive_constant)` is always `>= 0`, but is not able to do the same for `rem_euclid(x, positive_y)`.

Start with the Godbolt reproducer and compare it with the Alive2 proof to identify the missing deduction for a variable positive divisor. Trace where rem_euclid is represented and optimized in Clang/LLVM, then add or update a regression test showing that the result is proven non-negative for positive runtime values.

Written by the indexing model from the issue text.