Proposed Inclusions in gdual interface

For further numpy compatibility, the following improvements may be useful in expose_gdual.hpp in this section:

https://github.com/darioizzo/audi/blob/365a0719ec939319df57b792f996b0efea2a1225/pyaudi/expose_gdual.hpp#L145-L186

Proposed inclusions:

.def(
            "log10", [](const gdual<T> &d) { return log(d)/log(10); }, "Base 10 Logarithm.")
.def(
            "log2", [](const gdual<T> &d) { return log(d)/log(10); }, "Base 10 Logarithm.")
.def(
            "log1p", [](const gdual<T> &d) { return log(d + 1.0); }, "Logarithm of x+1, inverse of exp(x) - 1")
.def(
            "expm1", [](const gdual<T> &d) { return exp(d) - 1.0; }, "exp(x) - 1, inverse of log1p")
.def(
            "erfc", [](const gdual<T> &d) { return 1.0 - erf(d); }, "Error Function Complement")

A __float__ method for extracting the constant coefficient of the truncated taylor series en-masse. Useful for converting an array of gduals into an array of floats, convenient for certain computations and simplifies the process of extracting the constant part for large arrays. Can do this:
```
test = np.array([gdual_double(1.0, 'x{}'.format(i), 3) for i in range(1000)])
test_constant_cf = test.astype(np.float64)
```
Instead of
```
test = np.array([gdual_double(1.0, 'x{}'.format(i), 3) for i in range(1000)])
test_constant_cf = np.array([i.constant_cf for i in test])
```

Not sure how to implement these in an efficient manner, but they may be a good idea:

An arctan2 function that uses the constant part of gduals to determine the quadrant. Since the truncated Taylor series has a finite of convergence, then for values far from the axis lines, it is reasonable to assume that evaluating this expansion around some nominal point in a quadrant will give values in that quadrant. ie. arctan2(x, y) and arctan2(x+dx, y+dy) are likely to be in the same quadrant for x,y far from the axis lines.
A hypot function for computing the hypotenuse of a right angle triangle given the legs. As per here

Proposed revisions:

inverse trigonometric and hyperbolic functions are renamed from atan to arctan as numpy uses the arc-prefix instead of the simple a-prefix.

darioizzo / audi

Proposed Inclusions in gdual interface #50