Strongly prefer requiring 0 <= k <= n at first.  This is a programming language:  it will be applied to real problems, not to abstract proofs where some slop can be helpful in reducing the number of cases that need to be considered.

The Twitter fellow is certainly right that "0" is the clear answer to "how many 5-element subsets does have a 4-element set have?", but in the context of real problems it's more likely (to my eyes) that a programmer asking that question of `comb()` has gotten confused.  It certainly would have been "an error" in any programming use for `comb()` I've ever had.

Raymond, I'm not clear on what you mean by "alias".  You mean supplying the same function under more than one name?  I'd be -1 on that (where would it end?  the answer to every future name bikeshedding issue then would be "all of the above").  But +1 on, e.g., adding a doc index entry for "binomial" pointing to "comb".