Fix AssemblyName cache hash and key

[sdmaclea]

Fixes #24135

• AssemblyName::Hash(DWORD dwIncludeFlags) was including should be ignored attributes in its hash.
• AssemblyNative::Load was using stackMark in a case where it wasn't needed.
• Compiler started erroring because RuntimeAssembly.m_assembly was unsused in C#. It is used in native code. Disable the warning.
• Add ContextualReflection LoadWithPartialName test case.
• Remove unnecessary NoInlining attribute
• Disable broken CoreFX test case Fix DefaultLoadContextTest corefx#37071

[jkotas]

src/System/Reflection/Assembly.CoreCLR.cs(40,20): error CS0246: The type or namespace name 'FileNotFoundException' could not be found (are you missing a using directive or an assembly reference?)

[jkotas]

Do you plan to add any tests for this?

[sdmaclea]

Do you plan to add any tests for this?

The Assembly.LoadWithPartialName test is completely missing from the ContextualReflection test. I will add it.

I expected the ContextualReflections test to catch the Assembly.Load(string) failure, but it didn't. The corefx rewrite of the test did however. I need to look closer at it, to figure out why this use case was missed. If/when I can repro, I'll edit/add the test.

I am adding tests in corefx. These found the issue, but since they are not yet passing they are disabled until these bits make it.

[sdmaclea]

Need to diagnose the root cause further. This is not fixing the observed issue.

[sdmaclea]

Spent a lot of time in debugger root causing this:

• The binder cache was not finding assemblies already loaded because the key lookup was failing.
• The AssemblyName Hash function was using the public_key and public_token when it wasn't supposed to.
• Also we were populating the parent assembly using the stack mark when we didn't need to. This led to cases where the wrong binder_id was used as the key.

I have added the LoadWithPartialName test. The corefx test will catch the binder cache issue. I used that to debug the failure. I should be able to create a lighter test, but I'll put it in a separate PR.

I would like this fix to make preview5.

[sdmaclea]

There are two "regressions" in the test runs.
https://github.com/dotnet/corefx/blob/13cd96122c25328a1b180a6289923657180a7b69/src/System.Runtime/tests/System/ActivatorTests.netcoreapp.cs#L263-L265, Is testing a resolve event is fired when the assemblyName is illegal/malformed. That is the removed CreateAssemblyName function. I'll remove that change.

The other regression is.
https://github.com/dotnet/corefx/blob/13cd96122c25328a1b180a6289923657180a7b69/src/System.Runtime.Loader/tests/DefaultContext/DefaultLoadContextTest.cs#L260-L270

The test is bad. It is regressing because the assembly cache is now finding the System.Runtime assembly which was loaded before clearing the LoadedFromContext flag. So this assert is failing. Moving the clear earlier in the function seems to resolve the failure.

[sdmaclea]

This looks like it will miss preview5.

[jkotas]

resolve event is fired when the assemblyName is illegal/malformed

FWIW, I would not mind losing this "feature". Illegal/malformed assembly names will fail everywhere else, except this one place that tries to recover from them. It can be done separately / later.

[jkotas]

Thanks!