Improvement request for the local testing script for microbenchmarks

[fanyang-mono]

To summarize the suggestion listed in #3338. I would like to propose several enhancements for consideration to be incorporated into the local testing script for microbenchmarks:

• Remove Killing any running dotnet, vstest, or msbuild processes, when running on developer's local machine (Add option --dont-kill-dotnet-processes to benchmarks_local.py #3431)
• The script will have an option to build a docker container. Run the microbenchmarks there and report back the results. - Improvement request for the local testing script for microbenchmarks #3385 (comment)
• When running with admin privileges, it would be nice that only the microbenchmarks are run with higher privileges and building of runtime is done with standard user privileges. - Improvement request for the local testing script for microbenchmarks #3385 (comment)
• Add the profiling option, specifically, I would like an option to use EventPipeProfiler. The document for it could be found here This will help use trace down c# code path change, when investigating a regression. (Answer: Improvement request for the local testing script for microbenchmarks #3385 (comment))
• Add a proper README for the script. Add initial README for the benchmarks_local.py script. #3468
• Add filter aliases or similar for microbenchmark sets that are often problematic. (per below) (Initial non-integrated location in the README: Add initial README for the benchmarks_local.py script. #3468) (Feel free to add your own!)

[LoopedBard3]

For the profiling option, adding --bdn-arguments "-p EP" will enable that specific profiler. Any BDN specific arguments can also be passed this way, such as --buildTimeout. Since this is not documented great, I will make sure to add a section on it to a proper README (I also added creating a README to the list of items).

[matouskozak]

I was trying the local benchmark script on DevBox x64 WSL Ubuntu and it takes quite a while to run all microbenchmarks (10-13h depending on the configuration). I was thinking, if for local testing, we might disable some long running microbenchmarks or try to select a set of microbenchmarks that usually are the problematic ones (on Mono side those are vectors, strings, span, collections) and create an alias for the --filter argument to run just those.

[LoopedBard3]

I have a PR out for an initial README: #3468. All feedback is welcome. Right now, there is a placeholder section for useful filters that could also use some suggestions.

[LoopedBard3]

With the merging of #3934, which removes the default of terminating all dotnet processes during runs, a major reason for having containers to run the script in is fixed.

For the building of the runtime as non-admin, I was not able to find any clear path to achieve this cleanly for both windows and linux in python. A manual alternative to this is to initially only build the runtime with --build-only and --allow-non-admin-execution arguments set, and then run again without them set. Unfortunately, this would require two calls to the program. If a clear path is found that enables this work, then it can be revisited.

With all of this, I am closing this issue. Please leave any additional comments/feedback either here or in a new issue.