Prepare dataset for SFH and mass history as tf dataset

[yannick1974]

There is still some problem of some CSV SFH missing some SnapNUm, e.g. TNG100_mainprojenitors_102694.csv is missing SnapNUm 13.

[nicolas-cerardi]

Ok, so I see 3 options :

1. We write 0 anywhere a SnapNum is missing, and make a boolean feature for to tell wether the timestep has been inserted or not. So imagine t = [1, 3], d=[2.5, 1.5], then the dataset sequence will be : t = [0, 1, 2, 3], d=[0, 2.5, 0, 1.5], b=[0., 1., 0., 1.]
2. We write as many 0 we need to get N=100 before the oldest SnapNum. In this case we can still use N_age and don't need a boolean sequence. t = [1, 3], d=[2.5, 1.5] => t = [0, 0, 1, 3], d=[0, 0, 2.5, 1.5], N_age=2
3. + as suggested by Marc, interpolation (if there is only 1 missing value)