Work out the design for the virtual object initialization API

[FUDCo]

In #1960 we've been having a debate about the proper design of the API for initializing virtual objects (please see that issue for the discussion thread leading up to this). However, that issue was actually just a PR for some updates to the documentation, hence not really the right place to try to resolve the API design question. The correct issue to do that is right here.

[FUDCo]

@erights @warner @michaelfig

[FUDCo]

To summarize: @erights is displeased with two aspects of the current design: (1) contamination of the developer's namespace for methods on their virtual objects by having initialization flow through a special method named initialize, and (2) magically tinkering with the object that defines the virtual object's behavior by deleting the special initialize method from it so that it won't be exposed to holders of references to the virtual object.

Mark, am I fairly summarizing your concerns?

[FUDCo]

In a conversation with @warner this afternoon, he suggested we might mitigate the namespace concern by keying the initialization method with a symbol available from the makeKind function, e.g., [makeKind.initialize] or some such. This would not address the concern about having a magic method, but wouldn't be a name that a virtual object developer would want to otherwise use.

[FUDCo]

Closed by #1967