Add `categorical` vector support to `randsample`

[SergeantQuickscoper]

Currently, randsample.m requires categorical vector support for input population argument to be MATLAB compatible, as observed in the docs. This issue proposes adding support and necessary tests to randsample.m as part of the ongoing datatypes integration effort.

Current categorical input seems to have this behavior:

octave:24> C = categorical({'red','blue','green','red','blue','green'});
octave:25> C
C =
  1x6 categorical array

    red    blue    green    red    blue    green    

octave:26> y = randsample(C, 1)
y =
  categorical

   red
octave:27> y = randsample(C, 2)
error: randsample: The input k must be a non-negativeinteger. Sampling without replacement needs k <= n.
error: called from
    randsample at line 54 column 5

where any k > 1 will raise an error.

Note: string support seems to already be in place, but I did not test extensively. Could also look at that while tackling this issue.

[SergeantQuickscoper]

Well, the issue seems to come from numel always returning 1 for categorical vectors (always returns 1) in octave. This seems to be by design from datatypes in categorical.m:

## For compatibility reasons with Octave's OOP interface and @code{subsasgn}
    ## behavior, categorical's @code{numel} is defined to always return 1.
    ##
    ## @end deftypefn
    function out = numel (this, varargin)
      out = 1;
    endfunction

Anyways, luckily as population inputs must be vectors, using length instead should resolve this.