categorical sampler has length control parameter

[danielturek]

The title says it all.

Following suggestions in NCT # 505.

[danielturek]

Specifically, this addresses using the categorical sampler on distributions other than dcat. In such case, the issue is to determine the number of possible outcomes (or categories) k, where the distribution outcomes are the values $(1, 2, 3, \ldots, k).$

In support of that, this PR:

• Adds a length control parameter to the categorical sampler.
  • If length is a character string, it gives the distribution parameter name, for which the length of this parameter defines the value of k.
  • It length is numeric, it itself defines the value of k.
• The check for a dcat distribution is relaxed. A second control parameter anyDistributionOK can be set to TRUE to subvert this check.

[paciorek]

@danielturek I used the control element named check to be able to turn off some checking in the (new) PG sampler. Maybe we should use that instead of anyDistributionOK. I can change it if you're ok with that.

[danielturek]

@paciorek Change just made on a branch, with PR, please merge it if it looks good.