Parallelisation and improvement of adapt_grid()

[avigan]

I propose several improvements for adapt_grid():

• parallelisation using multiprocessing module and the use of apply_async()
• creation of the spectro and photo grids without any assumptions on the number of parameters
• fix for issue Conditions never met adapt_grid() #5 using np.any()
• use of tqdm instead of the manual computation and formatting of time left

Note that the parallelisation is not super efficient, probably because the underlying functions to adapt the models would require some work to better in a parallel environment, but I still get a factor ~2 improvement on my laptop.

I have checked that the outputs are identical to the current version of ForMoSA in branch activ_dev, except on grid points that were not previously caught as bad points (see issue #5).