Interposition is redundant to interpretation

[robrix]

InterposeC intercepts requests for an effect without extending the signature; instead, it projects out and handles requests for that effect, passing any others along.

InterpretC and other typical carriers extend the signature at the left, and handle any requests for their signature, passing any others along.

Both extend the monad stack with another layer.

From the perspective of code run inside the monad, it makes no difference whether the available effects are A :+: B :+: C and B is interposed, or B :+: A :+: B :+: C and requests for B are intercepted by the leftmost carrier for B. (It is technically possible to observe the difference since you can constrain the signature and pattern match on the effect sum with a carrier, but there’s no point to doing so and so you simply don’t.)

Therefore, interposition as currently implemented is redundant: you could just have a carrier, either using runInterpret or manually defined, which extends the signature. Doing so would be quite useful since InterposeC is inefficient and complicated, and it would also allow us to remove prj altogether.

[robrix]

I propose:

1. Removing InterposeC, and suggest InterpretC in the changelog.
2. Removing Project/prj.
3. Removing the custom interposition tests which are anyways kind of redundant to InterposeC itself.