Deduce Approx epsilon for float

[philsquared]

Approx() takes its arguments as doubles - which is fine, except when one of them is a float and it picks up the epsilon value of a double.

Taking the arguments generically, and using epsilon for float when the argument is a float, should address the issue - even if the values are then stored as doubles.

[philsquared]

epsilon/ tolerance should probably be std::numeric_limits::epsilon() * 100

[philsquared]

Left Approx non-generic, but use epsilon for float (*100)