Only track the latest PartitionState for each partition, plus the initial attempt time.

[danielgmyers]

This change reduces memory allocations when building the WorkflowState object, particularly when a step had a large number of retries. This was accomplished in two ways:

• The SDK EventType.eventType() method allocates memory every time it is called, to generate an enum object from the source string. This method was being called a lot while reading through the workflow event history. Now, the method is generally only called once per event.
• WorkflowState now only builds one PartitionState object per partition, rather than one per retry attempt.

For a workflow that had a step with 1000 retries, these changes reduce memory allocations during WorkflowState.build() by about 75%.

I also moved the workflow cancellation handling into one place instead of being split across three different methods.

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[danielgmyers]

Someone pointed out to me that a newer version of the AWS SDK fixes the enum issue, so I'm reverting the eventType() related changes in favor of just updating our SWF SDK to a newer version.