Fix JsonReader and JsonTreeReader nextLong() not failing on precision loss

[Marcono1234]

Fixes #1727

JsonReader.nextLong() tries to parse a value as double in case parsing as long failed. An exception should be thrown if converting the double to long causes precision loss.
However, previously if the parsed value could not be represented exactly as double, the precision loss was not detected.
For example:

// Last digit is different but values are considered equal
Double.parseDouble("9223372036854775808") == 9223372036854775807L

This pull request changes it so if the number (or number string) represents a value which matches the syntax required by Long.parseLong(String), then the value is not parsed as double.
However, if the value does not match that syntax precision loss is still possible. But since the value uses double syntax, it is assumed that the source of this data is aware that floating point can cause precision loss. Trying to prevent this as well would require writing an own parsing method similar to Double.parseDouble to remain backward compatible and the amount of effort is likely not worth it.

This pull request also updates JsonTreeReader to fall back to parsing as double when parsing as int / long fails and to check for precision loss as well.