This PR fixes #3582 by defaulting [[CycleRoot]] to the module itself in case of error.

In case the module is effectively on a cycle, this would be lying, but in practice it doesn't matter: when a module has an error, the error is immediately propagated to all its ancestors, so the real cycle root has the same error state.

I originally tried to set [[CycleRoot]] properly, which requires more complexity. This code that sets the states of the modules in stack in case of error, needs to:

• first, iterate the stack reversed, propagating setting [[DFSAncestorIndex]] of each module to min([[DFSAncestorIndex]] of module, [[DFSAncestorIndex]] of its child)
• then, when iterating in order, it needs to set [[CycleRoot]] to the module itself unless the [[DFSAncestorIndex]] is the same as the previous module, in which case it copies the [[CycleRoot]] from it.

While this complex approach is more correct, I discarded it because it's still not 100% correct. In this graph, assuming that E throws, we will have not traversed F at that point so we don't know that D should actually have [[CycleRoot]] set to C.

In https://github.com/nicolo-ribaudo/ecma262/tree/cycleroot-fix-ugly I have an alternative fix if you prefer, that instead of ensuring that [[CycleRoot]] is always set it acknowledges that in this specific case it will be ~empty~.