Unmanaged struct keep-alive logic for async mount

[LTRData]

• Implemented manual marshalling of structures and wrapped them and original managed objects in SafeBuffer derived classes to keep managed data alive while the native structure is in use.
• DOKAN_OPTIONS and DOKAN_OPERATIONS changed to classes instead of structs so that they do not need as much copying around and it also makes the "keep-alive" logic simpler.
• IDisposable DokanInstance with a Dispose method that cleans up native allocations.
• DokanHandle class that implements SafeHandle logic for Dokan handles.
• Dokan.CloseFileSystem() removed and replaced with DokanInstance.Dispose().
• Added launchSettings.json to .gitignore because that file contains settings that each developer mostly want to have their own settings. It is only used when debugging in Visual Studio.
• Had to remove netstandard1.3 support to make things simpler. I assume netstandard2.0 is old enough these days anyway.
• Added netcoreapp3.1 target framework for test project and some if's around parts that are .NET Framework specific. This makes it easier to test netstandard2.0 library.

[Liryna]

Thanks @LTRData ! Very appreciated that you looked into this so quickly!!!

[Liryna]

This PR looks fine. The test issue comes from the overlapped tests which one of them assert due to unexpected data. dotnet test --filter OverlappedTests Otherwise all other test pass.

Thanks @LTRData for your PR! Merging it and I will try to find out what is happening.

[LTRData]

Thanks! There are some obvious errors in the tests. For example a handle could be closed before I/O is complete in some cases. Correcting that helped a bit but not completely. There is another issue somewhere that I cannot find the root cause for right now. The implemention passes other file system test cases I have tried so I am pretty sure there is nothing actually wrong, it is just the test cases in the solution that have some problems.

[Liryna]

Well the latest CI run just passed https://ci.appveyor.com/project/Liryna/dokan-dotnet/builds/42134826 ... even if locally I still have failures 😶‍🌫️ I have been debugging the kernel to see if all data were completed correctly and it is the case. So Yeah I am on your side that something is wrong with the test but I am not able to point what.

Correcting that helped a bit but not completely

Could you share the change ?

[Liryna]

Oh just found! https://github.com/dokan-dev/dokan-dotnet/commit/4456212fbd536f658cafa346bf478e4b408a0c97 Have you seen any other issues ? It is still flaky for me

[LTRData]

I have not found anything else that looks like it could have any real impact on anything, but the code is a bit difficult to follow. It could probably be worth it to try to write entirely new test cases for overlapped tests. In earlier versions the threading was different in the driver and so I guess it probably happened to work even though something was probably broken with the test cases already.