ENH use variadic templates args in scatter and figure

[Bidibulke]

In order to avoid adding other declaration function for the same functionnality with other arguments, I had the possibility to pass extra-arguments in a tuple of variadic templates.

The kwargs of matplotlib functions are accessible using a tuple of pair (string, T). T can be :

• bool
• int
• long
• double
• char*
• string
• vector of double
• vector of int
• vector of vector of double

The syntax is not heavy, and if other types are needed, just implement a function such as

std::pair<std::string, PyObject*> analyze_key_value(const char* key, const T& value) {
  // Transform object of type T into an python equivalent type
  PyObject* pyobj = ....;
  return {key, pyobj};
}

Some examples:

• streamplot

vector<vector<double>> lw;
// some code to fill lw
plt::streamplot(x, y, u, v, make_tuple("color", mag, "linewidth", lw));

• scatter

// z is a vector<double>
plt::scatter(x, y,
                 make_tuple(
                                     "marker", "o",
                                      "c", z,
                                       "s", z,
                                      "cmap", "inferno",
                                      "alpha", 0.5
               ));

[Bidibulke]

I added streamplot function (still with variadic template args). I also extended quiver to accept a vector color as 5th arguments (in the generic form), and new tests for those functionalities.

[Bidibulke]

• New functions: plot_surface, imshow and stem.
• Tests with Catch2 have been updated.