Big performance regression with allocations for .TryParseUnit

[carlosesquivelroisenvit]

Describe the bug
We have found a big difference in performance between versions 5.74 and 4.152 when doing XXX.TryParseUnit for large datasets. Version 5 allocates a lot of memory when parsing the units versus the previous version.

To Reproduce
Attaching small repro projects with BenchmarkDotNet

Expected behavior
Performance should be comparable between versions

Screenshots
Adding BenchmarkDotNet output for both versions

Version 5

Method	ArraySize	Mean	Error	StdDev	Gen0	Gen1	Allocated
ConvertInchesToMm	400000	5.962 s	0.1036 s	0.1063 s	607000.0000	6000.0000	10.64 GB

Version 4

Method	ArraySize	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
ConvertInchesToMm	400000	76.77 ms	1.474 ms	1.379 ms	16571.4286	285.7143	285.7143	296.03 MB

BenchmarkTests.zip

[angularsen]

Could you give the latest pre-release of 6.x a try? I believe there are some relevant allocation improvements there.

[carlosesquivelroisenvit]

Hi @angularsen, just did with 6.0.0-pre014. Results are roughly the same than for 5.74

Method	ArraySize	Mean	Error	StdDev	Gen0	Gen1	Allocated
ConvertInchesToMm	400000	5.873 s	0.0425 s	0.0332 s	622000.0000	6000.0000	10.92 GB