Implement strict sign for r2d2m2cp

[ferrine]

There are cases where variable sign may cause symmetries in the posterior distribution, but the prior is still usefull in the context

So far

sigma, beta = pmx.distributions.R2D2M2CP(
            "beta",
            1,
            inputs.std(),
            dims="vars",
            r2=0.6,
            r2_std=0.1,
            positive_probs=[1, 0.7],
            variables_importance=[20, 10],
            centered=True,
        )

Will fail because nan will be resulted due to positive_probs[0] = 1

In this case I think it does make sense to use another distribution for the prior. The immediate choice is the Truncated normal distibuton with mean ans sigma calculated at p=0.99.

Additionally, in case positive_probs_std is provided, std=0 would be required for the entry

Any other thoughts on this?

[ferrine]

the idea is to set the distribution of such coefficient to a deterministic value as std limits to zero