uniform_real_distribution should use uniform_int_distribution

Hello again!

`generate_uniform_real` for `is_integral == true` should probably do something like this (pseudo code):

auto dist = uniform_int_distribution( unsigned long min, unsigned long max );
auto numerator = dist(eng);
auto divisor = unsigned long max;
[...]

instead of using:

result_type numerator = static_cast(subtract()(eng(), (eng.min)()));
result_type divisor = static_cast(subtract()((eng.max)(), (eng.min)())) + 1;
[...]

The current implementation fails to produce more different values than the range of the Engine is able to give. E.g., a PRNG which only ever produces values in the range [0..1] would lead to returning only 2 different real values.

What are your thoughts about this?

Best regards,
Dominik

PS: I noticed this when using good ol' `rand()` as an Engine behind the scenes.