We can generate non-whitelisted structs using opaque items

In #452 I'm removing an assertion that was hitting in stylo, concretely with the BaseErrorResult typedef.

This codegen was being called from CompInfo::codegen as expected (it's an inner typedef). The above item was clearly whitelisted (it'd have hit before otherwise).

Upon reflection, ErrorResult is marked as opaque, so we immediately stop tracing it (never whitelist the inner item).

When we arrive to generate code for ErrorResult, we generate opaque structs, but we still generate the inner types for it (which is desirable IMO).

What should we do here @fitzgen? Tracing across opaque stuff doesn't seem desirable, but neither does not generating sub-items for opaque items. Presumably we could change the assertion to something like "is whitelisted or any parent is opaque", or something like that.

Start at CompInfo::codegen and trace how ErrorResult, its opaque status, and the inner BaseErrorResult typedef are handled. Reproduce the stylo case described in the issue, then inspect the whitelist assertion and related code-generation behavior. Done means the intended opaque-item handling is decided and non-whitelisted inner structs are handled consistently without the assertion failure.