Update clrmd to 3.1

[timcassell]

Fixes #2383

I left the x86/x64 disassemblers on v1, those can be updated separately.

Also, as discussed in #2383, this breaks netcoreapp3.0 and older, so we should probably release this in 0.14.0.

[adamsitnik]

I am supportive of this change, but we need to perform a LOT of testing.

Thank you for your hard work on this @timcassell !

[adamsitnik]

It took me a lot of time to workaround all ClrMd bugs. If we really want to remove these workarounds we need to do a very extensive testing:

1. Run a simple benchmark and ask BDN to disassemble ALL methods for main. It should produce a markdown file with all CLR methods disassembled.
2. Same as above, but for this PR.
3. Diff two produced .md files and ensure that all instructions are correctly resolved and there are no missing things.

[timcassell]

@adamsitnik I ran dotnet run -c Release -f net8.0 --filter BenchmarkDotNet.Samples.IntroDisassembly.SumLocal --disasmFilter '*'
And it looks like we get a lot more information in 3.1 (as expected since clrmd was specifically updated to account for changes in the runtime).

I then ran the same benchmark with net6.0 to compare the older runtime results. The output was in a different order, but after comparing individual functions, they look identical.

[AndreyAkinshin]

I have tested this branch with some benchmarks, and it works like a charm! No concerns from my side: I haven't found any issues. However, I'm not aware of all the corner cases and I don't know how to perform proper extensive testing. I approve the PR and I don't mind including it in v0.14.0.

@adamsitnik what do you think? If you feel confident, we can merge it. If you have any questions, please provide a set of scenarios that should be tested.

P.S. I feel like it's just impossible to verify such changes properly. There is always a chance of getting a corner case with a regression. We can always just give it a try: release the new version, collect feedback, and fix any further issues once they are discovered.