Fix an incorrect check in TypedIntegerImpl

This patch fixes an incorrect check in the constructor of
`TypedIntegerImpl`. When `I` is a signed integer and `T` is
an unsigned integer, `std::numeric_limits<I>::min()` is negative
while `std::numeric_limits<T>::max()` is an unsigned value, so
`std::numeric_limits<I>::min()` will be implicitly converted to
an unsigned value, which is not expected. This will also cause a
`signed/unsigned integer mismatch` warning when we build Dawn with
clang.

In this patch we fix this issue by requiring `rhs` to be non-negative
and always convert `std::numeric_limits<T>::max()` and
`std::numeric_limits<I>::min()` to be uint64_t when `I` is a signed
integer and `T` is an unsigned integer.

Bug: 436300898
Change-Id: I7deed22a5ae1a5a13d0a39e46a812f6bbb2216ec
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/259897
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Loko Kung <lokokung@google.com>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>