Investigate different thread allocation semantics under the hood

[farmdawgnation]

It's been reported that every time an Http is constructed, we get 2*$numCpus threads created by AHC because we create an AHC client on each invocation. Some things to look at:

• One option that's been suggested is to share an AHC client instance under the hood, which is thread-safe "with qualification" - specifically so long as you don't do anything not-thread-safe in your callbacks. The with qualification bit is what scares me about that.
• Perhaps a more targeted approach is to expose or better document a knob for the number of threads that we should allocate when we build the client.