Improve DFA matcher builder startup perf with many constraints

[JamesNK]

Fixes #46154

• Adds a cache to the matcher builder to reuse constraints marked as cachable.
• RegexRouteConstraint creates its Regex instance lazily. Still important in situations where the web app has many non-duplicate regexes.

[JamesNK]

Numbers:

Main (interpreted regex)

| Method | RegexSame |     Mean |     Error |    StdDev |  Op/s |   Gen 0 |  Gen 1 | Gen 2 | Allocated |
|------- |---------- |---------:|----------:|----------:|------:|--------:|-------:|------:|----------:|
|  Build |     False | 6.739 ms | 0.0801 ms | 0.0710 ms | 148.4 | 15.6250 |      - |     - |      7 MB |
|  Build |      True | 6.247 ms | 0.0859 ms | 0.0804 ms | 160.1 | 15.6250 | 7.8125 |     - |      7 MB |

Main (compiled regex)

| Method | RegexSame |     Mean |    Error |   StdDev |  Op/s |   Gen 0 |   Gen 1 | Gen 2 | Allocated |
|------- |---------- |---------:|---------:|---------:|------:|--------:|--------:|------:|----------:|
|  Build |     False | 24.85 ms | 0.485 ms | 0.476 ms | 40.24 | 62.5000 | 31.2500 |     - |     28 MB |
|  Build |      True | 26.00 ms | 0.284 ms | 0.251 ms | 38.47 | 62.5000 | 31.2500 |     - |     26 MB |

PR

| Method | RegexSame |     Mean |     Error |    StdDev |  Op/s |  Gen 0 |  Gen 1 | Gen 2 | Allocated |
|------- |---------- |---------:|----------:|----------:|------:|-------:|-------:|------:|----------:|
|  Build |     False | 1.521 ms | 0.0238 ms | 0.0211 ms | 657.2 | 5.8594 | 1.9531 |     - |      2 MB |
|  Build |      True | 1.235 ms | 0.0157 ms | 0.0139 ms | 809.5 | 3.9063 |      - |     - |      1 MB |

[JamesNK]

Note: Nothing has functionally changed in this class. I made changes to it (enable nullable, split up methods) intending to use it, but found I didn't need to. I left the changes here as they're all positive.

[danmoseley]

I guess if a constraint is evaluated once, it's likely to be hit an arbitrary number of times over the life of the app?

If that wasn't the case one could imagine lazily instantiating as interpreted first, then counting until a threshold and swapping for compiled.

[davidfowl]

This is similar to what DI does, but it could be something we do after this change. We would need to make sure this happened in a place that could properly measure the "hit count". I'm not sure where that is.

[mgravell]

looks good; one question

[davidfowl]

Very clever

[ghost]

@JamesNK, this change will be considered for inclusion in the blog post for the release it'll ship in. Nice work!

Please ensure that the original comment in this thread contains a clear explanation of what the change does, why it's important (what problem does it solve?), and, if relevant, include things like code samples and/or performance numbers.

This content may not be exactly what goes into the blog post, but it will help the team putting together the announcement.

Thanks!

[dougbu]

@JamesNK sorry this missed branching for preview1 by just a few minutes. If you want it in that release, please do the needful.

[JamesNK]

Waiting for preview 2 is fine. Thanks for the heads up though.

[javiercn]

Looks great @JamesNK