[REGRESSION] Host can throw InvalidOperationException at any point in time

[lauxjpn]

There is currently a regression bug in the AltV.Net.Host implementation, that can lead to the following exception:

InvalidOperationException:
AssemblyLoadContext is unloading or was already unloaded.

As it turns out, the Host class only holds a weak reference to the load context in its static dictionary and no other objects are holding any references on it.

coreclr-module/api/AltV.Net.Host/Host.cs

Lines 100 to 103 in f97394f

	var resourceAssemblyLoadContext =
	new ResourceAssemblyLoadContext(resourceDllPath, resourcePath, resourceName, isCollectible);
	_loadContexts[resourceDllPath] = new WeakReference(resourceAssemblyLoadContext);

Therefore, the load context is eligible for garbage collection at any time!

This is exactly what happens in our case, when we load larger libraries with lots of dependencies in our ServerResource.OnStart() method.
It happens at different points in time, with different assemblies, but GC is always triggert at some point in the start process and an exception is thrown on the next try to load an assembly into the already finalized load context.

This is a regression. It should work on the master branch, because it does not use a WeakReference there.

I will provide a PR, that fixes this issue and should not leak any memory either.

This was reported to our repo as a possible .NET Core 3.0 incompatibility of our database provider, but turned out to be unrelated to that (PomeloFoundation/Pomelo.EntityFrameworkCore.MySql#825).

/cc @lsvMoretti

[FabianTerhorst]

Thanks, the pr is merged and everyone has confirmed that it's fixed for them.