Improve performance of registry.metrics()

[SerayaEryn]

This improves the performance of registry.metrics() by approximately 40%.
This is achieved by creating a specific formatter for each metric.

Here are the benchmark results:

### Summary:

⚠ registry ➭ getMetricsAsJSON#1 with 64 is 2.343% acceptably slower.
✓ registry ➭ getMetricsAsJSON#2 with 8 is 0.8804% faster.
⚠ registry ➭ getMetricsAsJSON#2 with 4 and 2 with 2 is 1.183% acceptably slower.
✓ registry ➭ getMetricsAsJSON#2 with 2 and 2 with 4 is 0.2041% faster.
✓ registry ➭ getMetricsAsJSON#6 with 2 is 0.7355% faster.
✓ registry ➭ metrics#1 with 64 is 40.57% faster.
✓ registry ➭ metrics#2 with 8 is 38.32% faster.
✓ registry ➭ metrics#2 with 4 and 2 with 2 is 41.24% faster.
✓ registry ➭ metrics#2 with 2 and 2 with 4 is 43.03% faster.
✓ registry ➭ metrics#6 with 2 is 41.95% faster.
✓ histogram ➭ observe#1 with 64 is 4.848% faster.
✓ histogram ➭ observe#2 with 8 is 3.919% faster.
⚠ histogram ➭ observe#2 with 4 and 2 with 2 is 0.5036% acceptably slower.
✓ histogram ➭ observe#2 with 2 and 2 with 4 is 1.179% faster.
✓ histogram ➭ observe#6 with 2 is 11.81% faster.
⚠ summary ➭ observe#1 with 64 is 2.889% acceptably slower.
⚠ summary ➭ observe#2 with 8 is 2.747% acceptably slower.
✓ summary ➭ observe#2 with 4 and 2 with 2 is 2.210% faster.
✓ summary ➭ observe#2 with 2 and 2 with 4 is 0.1158% faster.
⚠ summary ➭ observe#6 with 2 is 3.125% acceptably slower.

[zbjornson]

Sorry for letting this sit. A lot of similar changes (the extends Metric changes) landed recently but I think the perf changes are still useful. Would you be up for rebasing this on master please?

[doochik]

@zbjornson Hi!
I've rebased this PR #373

[zbjornson]

Thanks! will review that one this weekend.

[zbjornson]

Oops, I just realized the original author of this PR @SerayaEryn is not @doochik, who rebased it. doochik preserved the authorship of SerayaEryn's commit though so you'll still get attribution. @SerayaEryn alright with you?

[SerayaEryn]

Yes, it is alright.