noAutoDestroy should seamlessly continue

[rkuhn]

Currently, the following code will block for the next live state:

const machine = createMachineRunner(...)
await machine.peek()
for await (const state of machine.noAutoDestroy()) {
  ...
}

Instead, the state stored as a result of peek() should be immediately delivered.

Up for discussion: this method is useful for sharing a machine owned by a cache with the application, but it would be even easier if it returned a full machine instead of just the next() and peek() methods.

[Kelerchian]

@rkuhn Per the last version, that is already the case. I vaguely remember a bug that precisely affected this

[Kelerchian]

#85 is only an additional test case that asserts this particular case