Remove InputSession

[antiguru]

The type doesn't add much value besides offering a narrower interface to InputHandleCore than the type did previously. With this change, the public interface provided by the latter type matches InputSession, which eliminates the need for a proper type to encapsulate the InputSession.

Makes the InputHandleCore::next function private as we'd like all uses to consume all inputs, and removes a unused internal function to access the handle's pull counter.

[frankmcsherry]

So, had a read through the code (I didn't make it easy on us) and think the reason InputSession exists is to provide the #[must_use] decoration. I.e. it is a way to communicate "you need to call either for_each or for_each_time on this". Whether that is a valuable thing to insist on (it's almost surely a bug if you do not) in exchange for the noise around it is a diff question though. But I think that's why it exists.

Still pondering if this feature is worth it, though.

[antiguru]

[T]he reason InputSession exists is to provide the #[must_use] decoration.

I don't think the #[must_use] works as expected. What it does is require all callers of functions returning an InputSession to consume the object. I think the only place to obtain the type is through InputHandleCore::activate, so the annotation on the type has the same effect as specifying #[must_use] on activate. It does not affect Rust's behavior around consuming arguments to closures, which are locally-bound variables, so the whole "unused variable" machinery protects against not using the input parameter here.

[frankmcsherry]

Yeah, good point. I misunderstood what the attribute on a type means (rather than a function), but it seems to be the case. Linear types would do the right thing here, but this attribute does not do that.

[frankmcsherry]

Seems good!