IDataView Type system needs to be extensible to properly support image, date, and other types

[wschin]

To support loading Image from IDataView (as shown in #3263), user must modify ML.NET code, which means our type system is not extensive. Do we want to invest on the extensibility of ML.NET type system?

[TomFinley]

Perhaps we ought to change this to, the reflection based mechanism by which things like prediction engine specifically are not extensible to new types -- a problem that would not exist but for the fact that the DataView type system is extensible, which makes the title a bit misleading? What you think?

I also I have a possibly workable solution to that specific problem in #3263, let me know your reactions.

[CESARDELATORRE]

Agree. I think this is "a must" feature (pre-requisite) in ML.NET (support for in-memory images) in order to be able to release the TensorFlow and ONNX packages as v1.0 (Currently 0.12).

Note that this implementation would be centralized (not specific within the TensorFlow or ONNX packages), though, so it'll need a new version for ML.NET itself, too? Minor version?