Apply new technique for strong heterogenous array types

[briancavalier]

This uses a variation of the technique from Simpler Variadic Function Types in Typescript to improve the types of zipArray and combineArray for TS, allowing them to scale to any arity.

It also improves the type of mergeArray, allowing it to be heterogeneous while remaining strongly typed. Essentially, this extends the union behavior of merge to arbitrary arrays of streams. This works by leveraging the fact that TS union types obey a distributive law.

Potential Gotchas

In general, I see strong heterogeneous types as a plus. However, there could be a couple downsides:

1. They could be surprising or confusing for some users.
2. I haven't found any way to achieve parity in Flow. What are the implications for documentation?

Todos

• ToStreamsArray is duplicated, consider extracting it somewhere
• Figure out the implications for docs, given that Flow can't handle this (afaict)

[Frikki]

Consider using the newer readonly syntax for read-only arrays.

[unscriptable]

I’m going to try to find some time to investigate this technique in Flow. I’ve been looking for a generic solution for this pattern for a while.

[briancavalier]

I'd love to find at least an approximation in Flow. I tried (very) briefly to get something working with type-level functions (via $Call), but never managed to find something I was really happy with. It'll be great to have someone else give it a go!

[briancavalier]

@unscriptable Here's the gist of what I was fiddling with. Maybe it'll be helpful.

[briancavalier]

@unscriptable This isn't quite the same as the ZipArray case (which I still don't know how to do in Flow), but it's a useful technique nonetheless. It basically turns an array type into a union of the array's value types (removing the extra Stream nesting in this case).

[briancavalier]

Figure out the implications for docs, given that Flow can't handle this (afaict)

After looking at the docs, I'm inclined just to go with them as is, and let the Flow limitations manifest themselves as type errors for now. IOW, let's see if the current Flow arity limitations are problem in reality before we try to figure out what to do about it.