JSValue conversion to double precision floating point gives wrong results

In jsvalue.cpp we have:

  static int64_t ToInt64(double value) noexcept {
    return (std::numeric_limits<int64_t>::min() <= value && value <= std::numeric_limits<int64_t>::max())
        ? static_cast<int64_t>(value)
        : 0;
  }

However this seems wrong as the comparison between value and std::numeric_limits<int64_t>::max() is made by promoting max to double, which then loses precision:
https://godbolt.org/z/Ehxqef

int main(int argc, char**argv) {
std::cout << "int64 max = " << std::numeric_limits<int64_t>::max() << std::endl;
const double d = (double)(std::numeric_limits<int64_t>::max() - 511);
std::cout << "int64 max as double = " << std::setprecision(64) << d << std::endl;
 
return 0;

int64 max = 9223372036854775807
int64 max as double = 9223372036854775808

In fact, anything within +/- 511 of int64 max will be coerced onto a double with a value of 2^63. This is because the IEEE double mantissa is 52 bits, so anything above 2^52 will carry precision loss and may compare incorrectly.

Should we be limiting the ToInt64 function to only consider values within +-2^52 as valid in order to prevent loss of precision when converting to double? Or does it not matter?

Start in jsvalue.cpp at ToInt64 and reproduce the boundary behavior described in the issue, including values near int64_t's maximum and the double precision limit. Determine the intended valid range and verify that the chosen behavior handles the reported conversion cases without precision-related misclassification.