Track JSON perf with source generation

[pranavkm]

No description provided.

[SteveSandersonMS]

Cool! What kind of numbers do we get with this?

[layomia]

Since these are benchmarks, is it worthwhile to also track the fastest perf here which would be:

using MemoryStream ms = new();
using Utf8JsonWriter writer = new(ms);
personJsonContext.Person.Serialize!(writer, largeOrgChart);
writer.Flush();
serializeValue = Encoding.UTF8.GetString(ms.ToArray());

[pranavkm]

In general, we'd like these to be good representation for what our users are doing. These benchmarks are much more useful for us to track relative performance improvements / regressions in these user scenarios rather than being absolute best-case numbers.

[layomia]

The Serialize method is public based on the assumption that some users would want absolute-best perf and avoid some overhead of the serializer. Granted, some serializer features like max-depth handling would not be provided in this model, but do we not expect a non-trivial amount of users to want best-case perf?

[layomia]

What kind of numbers do we get with this?

We won't see any perf improvement here yet since the serializer currently requires Strict number handling settings to allow fast-path logic to be used, while the benchmarks are specifying AllowReadingFromString via JsonSerializerDefaults.Web. This is not the correct check on the serializer side, since we really only need that WriteAsString is not specified. We'll need to fix this on the serializer side to see an improvement here - dotnet/runtime#53814.