Fix issue #267 - Reclone broken on windows

[Xeckt]

Hopefully closes #267

While I wanted to stick to fixing the issue at hand, it became apparent that the issue was with the mutex guards. While this issue could also occur on Linux, it was more unlikely due to the logical difference in their kernels on object handling. So the way we handled the Repository mutex was problematic.

What I hoped to achieve for full atomicity and cross-platform support is making sure all file handles are dropped regardless, and keeping hold of the mutex guard throughout the whole reclone process. It was quite difficult to think about how to do in this situation. I've had @TheKrol test the code and it worked for him so here's the breakdown:

• Acquire the mutex lock (or guard, whichever you prefer to name it)
• Make sure we catch if the mutex guard is poisoned or locked first, if so bail
• Clean and remake the temp repo path explicitly
• Clone into temp repo path, ready for replacement
• Create a dummy repository that won't be used nor have file handles at an OS level since nothing is accessing it, so we can use it to replace the mutex and force the OS to drop handles, making sure it works on all OS platforms
• drop(replace(&mut *lock, dummy)); - takes the old repo out of the mutex, puts dummy in it's place, returns it, then drops it from the mutex, will keep the mutex guard itself in tact so we keep hold of the thread while releasing all handles to the old repo we want to replace
• Rename, and move the newly recloned repo into the mutex lock

I wanted to make sure we kept hold of the mutex guard the whole time, as to achieve atomicity.

I'm still unsure if this is a good approach but it made sense to me, since the Repository struct isn't using tokio so we can't necessarily hold other threads while we work on it.

[zleyyij]

Not tested yet, just one point of discussion