Clarify if defined() should provide type narrowing.

[multimeric]

Originally brought up here: chanzuckerberg/miniwdl#720.

Basically, it would be intuitive for defined() to narrow a T? to T in the type system, and make this legal WDL:

Int? optional
Int x = if defined(optional) then optional else 0

However, I'm guessing that this isn't clarified because WDL doesn't have the idea of type guards as this is the only function of its kind.

I wonder what implementations should do here, and if the spec could clarify? This also might be a chance to consider if type guards could be useful.

[DavyCats]

I don't think making an optional implicitly non-optional because it happens to be in a if-then-else with defined is a good idea. I don't think it is the if-then-else's job to do this conversion and adding such condition depended behavior could be unexpected to the users. It also just feels inappropriate to add such situation specific logic to a piece of syntax that otherwise performs a very straight-forward task (ie. pick a value based on a condition).

Besides, WDL already has a function that achieves this exact effect: select_first

Int? optional
Int x = select_first([optional, 0])

[multimeric]

Yes, good point. And it's probably fair to avoid the complexity that comes with type narrowing.