Use standard C++ filesystem API only

[Klaim]

See rationale in libmamba/include/mamba/core/mamba_fs.hpp

[Klaim]

Here is where I am at:

• I fixed all conversion errors on my local build with visual studio;
• I didn't address yet the linking issues visible on current CI build, I don't see them locally
• For some reason I can't link a Release build because of symbols missing from fmt, I suspect it's because I'm not using the same compiler as the fmt's dll (I used VS2022 IDE instead of building from cli, so that's understandable);
• With the debug build, I get runtime errors coming from the YAML-CPP implementation, so it is opaque, but it might be due to the same mismatch of build than for fmt;
• I suspect that the path->string conversions will fail on runtime on Windows (with no experimental UTF-8 mode) whith unicode paths, but I didn't reach the point where I could test that yet. Hopefully the pytest mamba/tests/ should catch these. There are several solutions if it fails, we should discuss that once this at least build on all platforms.

[Klaim]

This version compiles and run on my Windows which is set to have UTF-8 by default in everywhere. However only the libmamba tests can be run, for some reason the pytest based tests all fail to find some executable.

[wolfv]

@Klaim it looks like one last Unicode test fails on mamba.

[Klaim]

@Klaim it looks like one last Unicode test fails on mamba.

Yes, it's the one revealing the need for wrapping the path type. I was doing that but I'm starting to think it might be more complicated than just using a std::string to_utf8(const fs::path& path) function that would do the work where we need it.
From what I tried, wrapping the type is a bit brittle because it means every time we use a free function or operator from std::filesystem::path we end up with that type instead of fs::path and that can lead to bad surprises.
I'm trying to figure out what would be best without disturbing too much of the code.

[Klaim]

(rebased over CI not reporting issues that should have been reported, which I expect to fail because I added a test that should fail before I will push more commits).

[Klaim]

OK Now the test fails as it should.
I will now smash the commits for the test together and add my solution (that cannot run locally)

[Klaim]

looks like my workaround for fs::removeand fs::remove_all when VS is old works (at least on the CI)

cleanup incoming

[wolfv]

Thanks for pushing this through until the end!