Reduce allocations

[cmeeren]

This library showed up as a hotspot in Azure Code Optimizations for my API.

I believe this library has room for peformance improvements. At least in terms of allocations (memory usage).

Below are benchmark results for a quick experiment I did, where I cached a function that parsed, validated, and formatted input phone numbers. I used 100.000 phone numbers. 10x means each number was reused 10 times (so 10.000 unique phone numbers), and 100x means 1.000 unique numbers reused 100 times each.

As you can see, even just at 10x the cached implementation vastly outperforms the uncached implementation, not only in terms of CPU (which is to be expected) but also in terms of allocations/memory. The latter is more surprising, given that caching normally gives lower CPU but higher memory usage, and this is what leads me to conclude that there is room for performance improvements in this library wrt. allocations.

Method	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Allocated	Alloc Ratio
Uncached_1x	199.25 ms	3.293 ms	3.080 ms	1.02	0.02	4000.0000	-	67.14 MB	1.00
Cached_1x	284.59 ms	5.493 ms	7.333 ms	1.45	0.04	6000.0000	3000.0000	100.93 MB	1.50
Uncached_10x	195.84 ms	3.186 ms	2.980 ms	1.00	0.02	4000.0000	-	67.14 MB	1.00
Cached_10x	34.98 ms	0.980 ms	2.781 ms	0.18	0.01	1000.0000	-	22.37 MB	0.33
Uncached_100x	199.36 ms	2.797 ms	2.479 ms	1.02	0.02	4000.0000	-	67.14 MB	1.00
Cached_100x	11.72 ms	1.095 ms	3.212 ms	0.06	0.02	-	-	13.89 MB	0.21