Support for inter-instance stack switching

[frank-emrich]

There is a currently a conceptual error in the implementation, where the chain of active stacks (= continuations + the main stack) is stored per Instance/VMContext.

This means that when a function $f calls an imported function $g, where $f and $g are not part of the same instance, their stack chains are completely separate. For example, it is not possible to suspend to a tag $t in $g and handle this in a resume block in $f (assuming that the underlying modules import and export the tag $t so that it is shared between the two).

This PR rectifies this situation by sharing a single StackChain object between all instances of the same Store. The VMContext then contains merely a pointer to this shared chain, rather than a chain of its own. This fully mirrors how the VMRuntimeLimits are already shared between all instances of a Store.

[dhil]

LGTM.

[dhil]

Did you test this on the baseline implementation? I am thinking it may just have worked because of the use of (global) thread-local state. Though, it should be easy to construct an example with host threads, which causes a segfault. But I suppose we don't care much about such things at this stage.

[frank-emrich]

Regarding the baseline implementation, I assumed that it would just work due to its usage of thread-local state. But I did make sure that the new test is actually executed for both implementations. Crossing threads would require performing a host call and then re-entering wasm. I'm not sure if the current wasmtime API actually allows you to carry a mutable Store across thread boundaries.
I guess that we do not overly care about that, but it's something I've discussed in #109