JitBench harness improvements (#15151)

JitBench harness improvements (#15151)

* Improvements to JitBench perf test

1) More of the scenario setup is delegated to the scripts in the JitBench repo. The run now consists of (a) acquire source (b) RunBenchmark.ps1 -SetupOnly (c) run dotnet MusicStore.dll over and over to get measurements.
2) A specific last known good framework version is used instead of an open-ended latest version. This should make the test more reliable day-to-day at the cost that we have to bump the version when breaking changes occur.
3) Added an option to test with a local copy of JitBench instead of acquiring one from github. This is useful to for testing simultaneous changes in both repos.
4) Updated the output parsing to account for the changes I made in MusicStore's console spew.

* Code review feedback

* Don't use precompiled views by default

In order to match the previous benchmark results we should avoid
using precompiled views.

* Fixed JitBench local repo issue

In my earlier change I accidentally broke the normal scenario when you do not use a local copy of JitBench repo. This case was failing with an exception in SetupStatics because the null directory didn't exist.