The Parsing of a Double value was always executing a `Long.parseLong(value)`, which generated a `NumberFormatException`.

Identifying that a Number is a Double or a Long can be easily achieve (in a naive way) looking for the decimal separator.

This simple change avoids the extra `NumberFormatException`

A simple JUnit test, parsing a `Long` or a `Double` 10K times shows the next values:

* Double (old parsing): ~42 ms
* Double (new parsing): ~6 ms
* Long (old parsing): ~7 ms
* Long (new parsing): ~7 ms

As we can see, the parsing for `Long` values stays the same (±1ms), while the parsing for `Double` is dramatically improved.

Reducing the number of exceptions also has a positive side effect in memory consumption.

The usage of `indexOf(char)` is slightly faster