Add metrics to track batches vs singletons in external queue

[armanbilge]

Currently, the external queue holds tasks in two styles:

• batches, which are enqueued when a local queue spills over
• singletons, which are enqueued when tasks are scheduled externally (i.e., from blocking threads or from other threadpools)

We want to better understand the makeup of the external pool, to optimize how and when the worker runloop interacts with it. One interesting change is that now that we have integrated timers and I/O into the WSTP, this has removed one of the major sources of singletons.

It also is an important metric for applications e.g. lots of batches may suggest that the system is overloaded—in an ideal world, local queues would never have to spill their work.