[mono][perf] Remaining AOT Linux x86-64 perf issues after RuntimeHelpers.CreateSpan<T>

[jandupej]

This issue tracks the performance issues that were not resolved or fully resolved by introducing RuntimeHelpers.CreateSpan<T> intrinsic to Mono. These were the original regression reports:
https://github.com/dotnet/perf-autofiling-issues/issues/12060
https://github.com/dotnet/perf-autofiling-issues/issues/12054
https://github.com/dotnet/perf-autofiling-issues/issues/11780

The following benchmarks were not improved:

• System.Tests.Perf_Int32.TryParseSpan
• System.Tests.Perf_Char.ToUpper('A', en-US)
• System.Tests.Perf_String.Interpolation_MultipleArgs('A', "Foo")
• System.Collections.Tests.Perf_PriorityQueue<String, String>.Enumerate
• System.Collections.IterateForEach<String>.ImmutableArray
• System.Tests.Perf_Decimal.Mod
• IfStatements.AndArray

The benchmark history from just before the regressions occurred to just after the fix can be found here Ubuntu x64 AOT

[kotlarmilos]

@jandupej Do we know why they regressed or investigation is needed?

[jandupej]

@jandupej Do we know why they regressed or investigation is needed?

I'll need to investigate.

[DrewScoggins]

Also the ToString methods in this issue are regressed still.

dotnet/perf-autofiling-issues#12121

[matouskozak]

These regressions has been resolved during .NET 9 development timeframe. I confirmed by manually going thru some of the regressions that the measurement returned back to original values or improved.