fmudesign: alternative to specify distributions with percentiles

[tralsos]

It would be nice if we could chose what the dist_param1 etc. should represent. Should it be minimum/maximum or P90/P10. This will require extra columns where one specifies this. One could also implement this as two distributions, e.g:

triangular that uses min, mode, max
triangular_p10_p90 that uses P90, mode and P10 as input
This request covers all continuous distributions.

[tralsos]

Merged with enhancement request on providing truncated lognormal distribution (we have truncated normal and lognormal currently). There is no wiki article on truncated lognormal and it is not mentioned in https://docs.scipy.org/doc/scipy/reference/stats.html. Neither in ERT.

Feedback from BKH: Possible implementation is to transform parameters to normal distribution, simulate a truncated normal distribution and then transform values back. Particularly useful if input is not mean and std but e.g. P90 and P10.
Often doing something similar for e.g. uncertainties in permeability where you want to multiply by 5 or divide by 5. Instead of having 1/5 and 5 as input, using pert distribution with p90/ml/p10 = -0.7, 0, 0.7. Then using 10^simulated values before exporting.