In my view, ind2rgb is a 'core' enough functionality that full compatibility could be obtained. I found an older bug report where I was suggesting a short fix:
https://savannah.gnu.org/bugs/?func=detailitem&item_id=41851#comment7
It fits with TMW description that "image values are clipped to the range [1, c]" and solves all the cases I came across without error or warning.

ind2rgb does not allow NaN/Inf values in the indexed image while Matlab behaves as if they were equal to the maximal index according to the size of the colormap.

>> rgb = ind2rgb ([NaN Inf 1], jet)
rgb(:,:,1) =
   0.500000000000000   0.500000000000000                   0
rgb(:,:,2) =
     0     0     0
rgb(:,:,3) =
                   0                   0   0.562500000000000

One possible fix would be to add the following before the input check:

if isfloat (x)
  x (isnan (x) | isinf (x)) = size (map, 1);
endif