Restore fast mesa display

[ederag]

This PR brings back ipynbplotutils.py that was removed by b2b785a.
ipynbplotutils.Plot3D has no interaction but is much faster
(2.5 s instead of 25 s with the new ipywidgets).

The image format has been changed from JPEG to PNG, initially to work around python-pillow/Pillow#2609,
but PNG should be very good for that type of images (lot of pixels with the same color) anyway.
By default the background is opaque black, as before.
Since PNG handles transparency, now a new background keyword argument is allowed,
e.g. to get a transparent background:

from pyoptools.all import *
from pyoptools.gui.ipynbplotutils import Plot3D
# from the tutorial
L = library.Edmund.get("32475")
SEN = CCD(size=(20,20))
R = parallel_beam_c(origin=(0.0, 0.0, 0.0), direction=(0.0, 0.0, 0.0),
                  size=(10, 10), num_rays=(5, 5), wavelength=0.58929)
S = System(complist=[(L,(0,0,50),(0,0,0)),(SEN,(0,0,150),(0,0,0))],n=1)
Plot3D(S, rot=[(0, pi/2, 0), (-pi/2, -1e-2, 0)],
       center=(0, 0, 50), size=(30, 30), scale=6,
       background=(0, 0, 0, 0))

From the docstring:

    background: (R, G, B, A) tuple of floating point values.
                All values must be between 0 and 1.
                A is the opacity (0 means completely transparent).
                Defaults to opaque black: (0, 0, 0, 1).

Tested with PyOpenGL-3.1.0.

[ederag]

With glDisable(GL_DEPTH_TEST) the lenses are smoother (no polygons with different shades as on the previous image):

And the material opacity was decreased to better see rays passing through.
Please let me know if you prefer these changes to be in a separate PR.

[ramezquitao]

Thanks for the contribution. I apologize for not merging it sooner, but I have been very busy. Next week I will check and apply all the PR