Reconsider sandbox evolution abstraction

[dblnz]

In practice, sandbox evolution seems to unify two fairly disparate things:

• the transition from an uninitialized sandbox to an initialized one that in practice is not customized
• the evolution of a MultiUseSandbox to have a new snapshot on the top of its stack.

We should probably worth think about whether or not this is worth keeping.

From the hello-world example:

• 1

    let mut uninitialized_sandbox = UninitializedSandbox::new(
        hyperlight_host::GuestBinary::FilePath(
            hyperlight_testing::simple_guest_as_string().unwrap(),
        ),
        None, // default configuration
        None, // default run options
        None, // default host print function
    )?;

    // Register a host functions
    fn sleep_5_secs() -> hyperlight_host::Result<()> {
        thread::sleep(std::time::Duration::from_secs(5));
        Ok(())
    }

    let host_function = Arc::new(Mutex::new(sleep_5_secs));
    host_function.register(&mut uninitialized_sandbox, "Sleep5Secs")?;

• 2

    // Initialize sandbox to be able to call host functions
    let mut multi_use_sandbox: MultiUseSandbox = uninitialized_sandbox.evolve(Noop::default())?

    // Call guest function
    let message = "Hello, World! I am executing inside of a VM :)\n".to_string();
    let result = multi_use_sandbox.call_guest_function_by_name(
        "PrintOutput", // function must be defined in the guest binary
        ReturnType::Int,
        Some(vec![ParameterValue::String(message.clone())]),
    );