The most consistent treatment would be to give it some value, other than -inf, which will propagate into NaN's in other calculations. Otherwise, outputs depend more on how denormals get handled rather than how Pd code is written. Outputs could vary from system to system and that's just buggy.

The factorial is generalized by the gamma function which has values at non-zero and non-integer negative values, but it likely doesn't have much practical use in Pd. You could write an external if it actually matters to you for a project.

So.... Since there's little point in being mathematically literal about it, it's fine to just choose a value (0 or 1) for values less than 1.  If it's me, I would choose 1, so that factorial calculations have the same value for bad inputs (values <1) as for the trivial input (=1).

Chuck