Stricter condition for type assignability

[RunDevelopment]

This adds a new assign function that determines whether a type T is assignable to an input with some definition type D.

The old condition was "T and D overlap." This condition served as well to prevent connects that could not be valid, but it's not enough anymore. The new Conditional node causes problems in that regard, because it allows users to create the union of 2 types A and B. So users can e.g. create the type Image | number which is assignable to a number input under the old condition.

To fix this issue, I modified the condition slightly. Instead of checking with T for overlap, it will now split T into its unioned types and check with those items. So the condition is now: split(T).every(item => overlap(item, D)). The key here is how T is split. It's a little complex, because the any type should still be assignable to everything. split essentially works like this:

function split(T) {
  if (T.isAny) return [T]
  const num = T & number
  const str = T & string
  const structs = T.iterateStructInstances()
  return [num, str, ...structs]
}

2 things to note:

1. The is-any check is a little fuzzy to allow types very close to any. This is necessary, because nodes can't actually output any. There is special handling for the Error type, so nodes will output at most any \ Error.
2. Iterating struct instances means grouping them by their definition. E.g. Image { width: 100 } | Image { width: 200 } | PyTorchModel will be iterated as [ Image { width: 100 } | Image { width: 200 }, PyTorchModel ].

The new assignability condition fixes the issues the Conditional node causes, but I expect that we will likely need to revise it further in the future.

Examples

Image or color:

Image or number: