Added choice of model_demo inside the run_simul function using the set of decisions rules based on :
* lambda (growth vs decline) for DD
* lambda and N0 (small vs large pop) for StochDemo

Updated other functions accrodingly

Updated plot_impact accordingly

Added imports (magrittr, popbio) to models M3 and M4 functions

Added piece "%>% sapply(min, 0.999)" in M3 and M4 to avoid survival parameter > 1.
Also added this in 'init_calib' function.

All models tested and work.

Made Model 3 (with DD) simulations faster by changing the calibrate_params function (based on optim) by an analytical approximation (using match_lam_delta function).

Tests reveals that the approx is largely good enough. Makes no differences in the results.

It works well but it's slow because of the optim procedure : lam_Nt > calibrate vr > Leslie matrix as f(Nt)