Add support for async worker processes

The idea is that you might be able to parallelize internally in the worker process, with the benefit of shared memory to cache details about dependencies. Async worker processes can receive & run multiple commands concurrently. They respond with results asynchronously in any order.

PR Notes:

• The DSL arg is solo_async=True because async is a python keyword.
• I'm sort of lukewarm on the idea of ListenableFutures - open to suggestions. The only real benefit is that it makes testing concurrent execution/cancellation easier.
• In theory the Sync worker pool can be cleaned up with the new interface (less of a need for a BurrowedWorkerProcess & lifecycle). I did not do this to keep this PR manageable.