Treat null, undefined, and NaN as viral contaigons

[kriskowal]

This adds NaN to the list of invalid inputs for operators, such that
they will all propagate as undefined.

[kriskowal]

I’m sure @mczepiel or @aadsm will want to have a chance to veto. Assigning to @mczepiel.

[mczepiel]

I don't have any particular use case in mind where I know it would have been nice to differentiate encountering undefined from NaN. It just seems like data we had that could be informative. Otherwise sanely stopping evaluation in the same manner as encountering undefined is reasonable.

[kriskowal]

Alternately, I could propagate NaN and use the value != null && value === value throughout FRB to detect and propagate invalid input directly to output. value != null being shorthand for value !== null && value !== void 0 WAT, and value === value being the only reliable way to distinguish NaN and only NaN SUPERWAT void 0 being the only reliable way to express undefined since undefined is a variable that can be, in fact, defined MEGAWAT and isNaN being unsatisfactory because it coerces strings to NaN before checking M-M-MONSTERWAT.

[kriskowal]

Oh, wait. That’s what I did.

[kriskowal]

Oh, no. This would just be a start. I could use this defined method throughout FRB internally and just pass the bad value forward. When I rebase this, I might expand on the comments with above.

[hthetiot]

@kriskowal could you rebase on master ?