Fix #13326: FP shiftNegative for template argument

[ludviggunne]

Would love to get som feedback on this since it's making quite bold assumptions about expression that depend on template arguments.

[chrchr-github]

Seems like more checks could use the new mechanism, e.g. zerodiv:

template <int b>
int f(int a) {
    if constexpr (b > 0) {
        return a / b;
    } else {
        return a - b;
    }
}

int g() {
    return f<0>(2)
}

[ludviggunne]

Seems like more checks could use the new mechanism, e.g. zerodiv:

Do you think that should be baked in to this PR?

[chrchr-github]

Seems like more checks could use the new mechanism, e.g. zerodiv:

Do you think that should be baked in to this PR?

Not necessarily, more like an idea for future work.

[ludviggunne]

Not necessarily, more like an idea for future work.

Ok. I suppose I can create a ticket for the more general case when/if this gets merged.