runtime: initial prototype to enable multi-value and reference-types

[nagisa]

These WASM extensions are enabled by default in 1.83 release of the Rust compiler.

We have discussed enabling these features between crt members and these are rough conclusions:

1. The enablement work is probably not super onerous, given that these features were already implemented in wasmer at the time near-vm forked off;
2. Similarly (to the best of my memory) these features have been implemented in finite-wasm;
3. However it is much more crucial to actually verify whether the new functionality does not expose new avenues that would permit non-linear compilation times, for example. With multi-value it is now possible for somebody to specify thousands of return values from a single function. Does the compilation (and execution) time remain linear with regards to the number of values here? Does finite-wasm actually handle this correctly too? We appreciated multi-value being disabled last time we were dealing with this class of problems in related areas (e.g. local initialization problem.)
4. Further, enabling reference-types requires enabling bulk-memory which introduces a new family of primitives for our system: memory.copy (memcpy), memory.fill (memset.) These are new in not only that they are "expensive" instructions that are executed in the host code and need their own fees, but in that their processing time is going to naturally depend on the arguments passed to the functions. We do have some infrastructure for this in finite-wasm (Aggregate instrumentation kind) but it hasn't been tested as rigorously as the other types of instrumentation kinds.
5. Old versions of the runtime (wasmer0) do not have an implementation for these features at all, however the choice of the runtime used and the validations performed are independent. We already have some precedent in panicking when there's an unsupported combination of the two, but at the same time we have many tests that validate "all types of the VMs." We will have to adjust these tests in a way that's not just disabling coverage for old VMs as changes in mildly related areas can break replayablility where these old versions matter.