Move the simulation thread pool from the model to the driver

[Twisol]

• Tickets addressed: N/A
• Review: By commit
• Merge strategy: Merge (no squash)

Description

This PR moves the ExecutionService for thread pooling from the model into the driver -- specifically, the SimulationEngine. Anything in the model that spawns a task now passes a TaskFactory back to the driver, rather than constructing a Task locally. This avoids a great deal of the work previously necessary to make the ExecutionService available wherever it was needed in the model, and generally streamlines interactions between the two sides.

As a result of these changes:

• RootModel is no longer necessary (as it simply bundle the ExecutionService alongside the mission's own model).
• The MerlinExtension no longer needs a model provided to it in test cases, so MerlinTextContext is no longer necessary. Also, the type parameters on MerlinExtension have gone away, meaning we can always use @ExtendWith(MerlinExtension.class) rather than the awkward field-based @RegisterExtension. Hurray!
• We can automatically use Loom's virtual threads if we detect that we're running in a JVM with them enabled. (We don't currently deploy images with --enable-preview present, but now it should be easier than ever to trial simulation on Loom.)

Verification

Build succeeds. I haven't actually tried enabling Loom on this PR because it's kind of a pain to configure Gradle, but if Loom isn't available it will fall back to the existing thread pool anyway. (I did actually try Loom just before JDK 19 was released, and it worked without hiccups. Anyway, if you run with --enable-preview, you're explicitly opting in to potential instability, so...)

Documentation

Probably some of the testing documentation needs to change. I also updated a code snippet for ActivityMapper that referenced RootModel when I was removing it.

Future work

• Move registration of resources and topics out of the model constructor. This will, as a side-effect, streamline tests even more, because we won't need to pass in a Registrar.
• Allow models to allocate cells during simulation instead of only at construction time. This would avoid the need for a test constructor entirely. Entirely! (But also, it opens up some interesting patterns for shared local simulation state between tasks that there isn't much reason to forbid.)