Switch to direct observation of costs on a per-operation basis

[TristonianJones]

Updates to the way costs are observed to reduce allocations an inefficiencies.

Benchmark	Master HEAD	Direct Push (Observerless)	Delta
simple_comparison	300.0 ns / 550 B / 10 allocs	205.1 ns / 224 B / 4 allocs	-31.6% latency, -59.3% memory, -60.0% allocs
function_calls	549.3 ns / 798 B / 14 allocs	355.6 ns / 288 B / 7 allocs	-35.3% latency, -63.9% memory, -50.0% allocs
comprehension	11,829 ns / 7,748 B / 148 allocs	7,583 ns / 5,188 B / 131 allocs	-35.9% latency, -33.0% memory, -11.5% allocs
nested_comprehensions	14,484 ns / 4,956 B / 250 allocs	7,357 ns / 3,340 B / 106 allocs	-49.2% latency (~2x faster), -32.6% memory, -57.6% allocs

[TristonianJones]

The fact that this wasn't 1 was actually a bug.