Add support for explicit type conversions from decimals, doubles and floats and rationals

[tompazourek]

Since all of these have only limited precision, it might be hard to get the right result.

E.g. number 0.33333333 will not get parsed as 1/3, which might be expected.

For this, maybe there could be a method, which will accept a certain tolerance coefficient. Then we can find the closest rational number approximation thanks to continued fraction expansions.