Add "requires binary" flag to opdef signatures.

[ss2165]

Operations in Extension declarations can state one of:

• An explicit type scheme
• A type scheme + a flag indicating binary required for validation, in-memory extensions should point to this function.
• A flag indicating code required to compute the type scheme, in-memory extensions should point to this function.

[acl-cqc]

How do I interpret this? Does this describe the current situation, or the desired state? If the latter, which bits are new?

Looking at the current op_def.rs, I note we use struct CustomValidator for every case where we have a PolyFuncType, even if there is no custom binary code - that seems undesirable, and the name is misleading (CustomValidator does not necessarily include any custom validation code!). I'd suggest:

pub enum SignatureFunc {
  //PolyFuncType(CustomValidator) // i.e., remove this, replace with:
  PolyFuncType(PolyFuncTypeRV, Option<Box<dyn ValidateTypeArgs>>),
 CustomFunc(Box<dyn CustomSignatureFunc>) // unchanged
}

However not clear how this relates to the issue here. A second possible goal is to replace each of the Box<dyn>s with an Option where None means "binary code require but absent", which is distinct from a None custom-validation (no validation required)....

[acl-cqc]

Ah, or does this just relate to serialization format?

[ss2165]

this relates to just serialised yeah, I've been thinking of those as "declarations" (as opposed to "definitions" that may be backed by executable code)