Fixing potential overflow issue in inner product trait

[lucbv]

When result type is double and inputs are floats, one input has to be cast to double so the multiplication operator for double is used instead of the float multiplication operator that could overflow for valid double values.

@csiefer2 this should take care of the security issue you reported.

[yasahi-hpc]

LGTM
Reasonable to avoid a potential overflow.

Oh, I have one question about complex overload.

KOKKOS_INLINE_FUNCTION void updateDot(Kokkos::complex<double>& sum, const Kokkos::complex<float> x,
                                      const Kokkos::complex<float> y) {
  const auto tmp = Kokkos::conj(x) * y;
  sum += Kokkos::complex<double>(tmp.real(), tmp.imag());
}

Do you need to apply casting for this case like

KOKKOS_INLINE_FUNCTION void updateDot(Kokkos::complex<double>& sum, const Kokkos::complex<float> x,
                                      const Kokkos::complex<float> y) {
  const auto tmp_x = Kokkos::complex<double>(x.real(), x.imag());
  const auto tmp_y = Kokkos::complex<double>(y.real(), y.imag());
  sum += Kokkos::conj(tmp_x) * tmp_y;
}