[BUG] String to Double return 0.0 for very large number and vice versa

[thirtiseven]

Describe the bug

cudf::strings::detail::stod returns inf for a number very close to zero, like 9.299999257686047e-0005603333574677677, and returns 0.0 for a very large number, like 9.299999257686047e0005603333574677677.

Steps/Code to reproduce bug

cudf::strings::detail::stod (link) uses int to save exponent value, so it will overflow to a negative sign if input exponent is larger than INT_MAX.
For the above 1st double string, it will overflow from negative value to positive value.
For the above 2nd double string, it will overflow from positive value to negative value.
So it gets wrong results.

  // check for exponent char
  int exp_ten  = 0;
  int exp_sign = 1;
  if (in_ptr < end) {
    char ch = *in_ptr++;
    if (ch == 'e' || ch == 'E') {
      if (in_ptr < end) {
        ch = *in_ptr;
        if (ch == '-' || ch == '+') {
          exp_sign = (ch == '-' ? -1 : 1);
          ++in_ptr;
        }
        while (in_ptr < end) {
          ch = *in_ptr++;
          if (ch < '0' || ch > '9') break;
          exp_ten = (exp_ten * 10) + (int)(ch - '0');  // Overflow here!
        }
      }
    }
  }

Not sure if this is intentional and matches some behavior.

getJsonObject in spark-rapids-jni needs this function to perform floating point number normalization, so we copied this function to spark-rapids-jni and fixed it. But we would like to call it from cuDF in a long term.