Improve in-proc consumption story of runtime metrics

[sywhang]

AB#1244363
Since .NET Core 3.0 we've been adding many DiagnosticCounters to the runtime as well as the BCL to track some key metrics. These metrics can easily be consumed by dotnet-counters or with ETW/EventPipe out-of-proc.

In-proc-consumption of these metrics is not very easy/efficient. Using EventListener to consume these metrics is non-intuitive and is also expensive due to the way EventListeners are implemented.

It's also not straightforward to get the metrics themselves by plumbing them through EventCounters -> EventSource -> EventListeners.

This issue tracks all the work that needs to happen to make in-proc consumption of runtime (and core libraries) metrics easier and more efficient.

Related issues:

• Provide managed APIs for our canonical set of runtime performance counters #36571
• A Real-Time Per-Thread/Task Perf Stats Tracker #43610

[josalem]

closed via #55046