Workers should be divided by spec, not fuzzing target

[jschwartzentruber]

Currently we allocate a TC worker-pool per fuzzing pool. We should instead define pools based on specs (machine types allowed, worker features/capabilities required), which can be used by one or more fuzzing pool. The max pool size would scale based on the sum of fuzzing pools using it.

This would allow us to modify machine specs without worrying about "stale" workers using an older configuration.

[jschwartzentruber]

Implementing this would also simplify scheduling on physical machines.