Surface module cache for reuse.

[graydon]

The goals of this PR are:

1. Surface the ModuleCache from inside soroban to its embedding environment.
2. Allow a module cache to outlive the soroban Host, and be reused from one Host to the next.
3. Allow the ModuleCache to be pre-populated from before the lifetime of a single Host, (eg. by stellar-core, with all the non-archived contracts currently live in the ledger, at startup or such.)

These goals are accomplished by the following structural changes:

1. Factoring out error-handling / error-mapping functionality of Host into a separate trait ErrorHandler that Host implements, and adding a new trait CompilationContext that extends both ErrorHandler + AsBudget. This abstracts the services the ModuleCache needs from the Host, such that the embedding environment can provide its own CompilationContext in the times pre- or post-Host lifetime.
2. Modifying the internal map that stores Modules inside the ModuleCache to use a refcounted-shared-mutable std::map::BtreeMap that can be incrementally and efficiently added-to or removed-from, and is not metered.
3. Module storage and arguments that were Rc<> are made Arc<> and the whole ModuleCache is made Send+Sync. This is mainly because we want to work with the reusable ModuleCache across C++ threads, and while Rust has no idea about C++ threads it at least helps us shoot ourselves in the foot less if all the Rust code that works with it preserves the sense of "this would be threadsafe in Rust if Rust was making the threads".
4. There is also a bit of gratuitous churn where things get renamed from foo => wasmi_foo and Foo => wasmi::Foo. This is because this change was done as part of (and there'll be a followup PR including) work equipping the Vm and ModuleCache with an experimental wasmtime/winch backend. It turns out this PR's work (on just the ModuleCache reuse part) is high enough value, and agnostic enough to which backend it's using, that we want to split this work out, move it up in the schedule and explore getting it into as early a release as possible. But I didn't bother reverting-out the wasmi_-qualification since it's intertwined with other changes here and harmless to keep.

This PR is a continuation of #1511 and should be merged after it.

This PR implements CAP-0065

[dmkozh]

Looks good to me overall, only not sure about the gating of the old behavior (left a comment about that).