Move to a common thread pool model

[AlexHarker]

At the moment contexts own thread pools. That's very inefficient in cases where lots of contexts exist.

Reworking the processing queue to remove the thread pool (putting that in the global) and making use of it from there would be an optimal approach. This has a number of issues:

• It becomes necessary to consider the number of items overall and per processing queue separately.
• It becomes necessary to have threads servicing multiple queues

This issue might also help with the memory locking discussed in #81 as it opens up the possibility of thread-specific memory pools.