fix(transaction): Fix schedule queues

[dranikpg]

If it reproduces locally on x86, it must be something simple. My only explanation is that we spuriously reorder the next read in sq.queue.Pop() before we update armed, so:

• T0: reads from an empty queue
• T1: stores the next context in the queue
• T1: makes sure armed is still true
• T0: clears armed flag and proceeds with stale data from (0)

[dranikpg]

But I still don't like schedule chains 😄 For one, they're overcomplicated, the armed flag & and a full blown mpsc queue is obsolete. For a second, it's only a partial implementation/fix

[romange]

I agree that their added value is questionable.