refactor(interpreter): invert frame composition ownership

[zhenrongliew]

Stacked on #684. Make each interpreter engine's private stack enum own conversion from reusable member frames.

Problem:

Concrete member frames depended on the total enum that stores them. BlockFrame required F: FrameBuild<V, E> and re-wrapped itself with F::from_block(self), so the dependency ran:

BlockFrame -> FrameBuild -> total enum -> BlockFrame
The cause was in the frame protocol itself. Frame<I, F>'s three methods all returned FrameEffect<F, C> — over F, never over Self. A reusable frame was therefore obliged to construct the enclosing enum's variant for itself, which it could only do through a construction trait the enum had to implement. That forced a chain of scaffolding: a FrameBuild derive per engine family, a Build* construction trait per dialect frame.

Fix:

Split the parent from the child in FrameEffect:

pub enum FrameEffect<P, C, F = P> {
    Continue(P),                    // P = this frame's own next state
    Push { parent: P, child: F },   // F = configured representation of a child
    Done,
    Complete(C),
}

pub trait Frame<I: FrameEngine, F = Self> {
    fn step_into(self, interp: &mut I) -> Result<FrameEffect<Self, Self::Completion, F>, I::Error>;
    // ...
}

A member continuation now returns itself for Continue and as the suspended parent of Push. A private closed stack-item enum is the composition root: it owns the From<Member> conversions, dispatches exhaustively, and lifts member state into its own variant with the new FrameEffect::map_next.

[planza-quera]

Is there a reason as to why the snapshot tests are no longer needed for this macro?

[zhenrongliew]

You're right, the deleted snapshot covered the old FrameBuild, AbstractFrameBuild, and DenseFrameBuild. I've not added new snapshots for the new frame.rs.

[zhenrongliew]

Added a snapshot covering member frame conversions, the new [derive(Frame)] macro. 6419a99