Observers can now take priority levels, which allow the user to control
the order of execution. Note that reactive expressions do not have
priority levels; since they are lazily evaluated, it wouldn't make any
sense to speak of priorities.
Another commit will be needed to add an API for changing the priorities
of outputs (probably in outputOptions?).
The first of the included tests did not pass without the changes to
Observable. The problem occurred when a function read a reactive value
and then wrote it. Any dependents on the function would not receive
any invalidations, then or ever after.
The first problem was that the dirty state was unilaterally set to FALSE
after the function finished executing, which might not be accurate if
the function's newly created was invalidated during its own execution.
Instead we set dirty state to FALSE before executing. But to prevent
reentrant calls from thinking the cached value can be used, we add
a .running field that is also consulted during getValue.
The second problem was that Observable$getValue didn't register the
dependent until after updateValue. That is a problem if updateValue
creates *and* invalidates a context before returning. So now we
register the dependent before calling updateValue.
