parameterization of continuous pdf

Continuous probability distribution functions can be useful. Often our probabilities are continuous and we discretize them without assessing the impact of this discretization on the evaluation of the problem.

copy of https://github.com/equinor/dot-archive/issues/189:

Is your feature request related to a problem? Please describe.
Both discrete and continuous pdf's may be needed in the decision model. While discrete pdf's can be parameterized by e.g. structured data such as (python) dictionaries, continuous pdf's are more problematic. In particular, we may not want to tie this parametrization to a given language and the use of any type of eval function should be prohibited for security reasons.

This issue aims at finding and implementing a relevant parametrization. some possibilities could be:

A list of predetermined pdf's such as Normal, Lognormal, Uniform, Poisson... (and their truncations versions) and a set of parameters defining the full distribution (eventually reusing some ERT/FMU functions)
Link to samples of the pdf (e.g. related to FMU/Everest)
Parsing a string describing the pdf, given a set of possible mathematical functions, such as **(or^) *, /. +. -. sqrt, exp, log, Heaviside, step (the two latest for truncation purpose)...

No files, tests, or entry points are named. Start by locating the decision-model representation and existing discrete PDF parameterization, then compare the listed approaches for continuous PDFs, including the security constraint against eval; done means a relevant parametrization is selected and implemented for continuous distributions.

Written by the indexing model from the issue text.