specify tolerance instead of max for distributions

[sbfnk]

Instead of specifying the maximum one could support specifying a tolerance, i.e. minimum pmf / 1-cmf as is already implemented in the convolution (or the conversion to stan code if going with the approach in #504). This might require looping over incremental values of max to find the "correct" one.

[seabbs]

Yes nice idea

[sbfnk]

As we can doing this in the conversion to stan code everywhere now, it is not a priority for v1.5.