Rather than spawning a new goroutine for each message,
run each log system in a dedicated goroutine.

Ensure that logging is still asynchronous by using a per-system buffer
(currently 500 messages). If it overflows all logging will hang,
but that's better than spawning indefinitely many goroutines.

Messages are formatted by generic part, so the log system doesn't need
to provide formatting. This fixes the test from the previous commit.

As a small bonus, log systems now have access to the level of
the message. This could be used to provide colored logging in the
future.

This test fails because the log message is formatted twice.

This ensures that examples will actually compile.

It's now safe for concurrent access.
Output checking looks better.

This also fixes a deadlock in the tests.

Blocking event package

If a state gets reset and you still hold a pointer to the previous,
incorrect, state object you'll operate on the wrong object. Using the
state to set/get objects and attributes you won't have this problem
since the state will always have the correct object.

* CALLVALUE pushed incorrect value to the stack
* Set execution model to closure