We see this assert triggered frequently, even sometimes in production,
so at least make it easier to track down the data involved.

This was triggering an assert error with vibe-d path.
The recursion logic was a bit murky, so the function
is now split in 3 different part and better documented.

This gets the unittest framework closer to the actual behavior,
and allows us to mock basic FS operations such as mkdir and writeFile.
Further improvements can be made to support more operations,
and to have package setup done before `TestDub` instantiation.

This was wrong on many level:
- The missing `else` clause meant that the `enforce` clause was always there on Windows;
- The `enforce("Reason")` should have been `enforce(false, "Reason")`,
  as in its current form it would never fail and not serve its intended purpose;
- It was pushing at runtime something that was easily detected at CT;

* Add missing attributes to NativePath.empty

* Dependency: Change format for path-based dependencies

So SumType infers its opEquals and co as safe.

Make sure a recent enough Vibe.d is used so we can update the vibecompat code.
Additionally, since the split of Vibe.d into package, and the splitting of
vibe-core into its own project, the version specification was never triggered,
so it was corrected in this commit (the version spec was correct for older Vibe.d).

There are plenty left, but those ones didn't seem to have any side effect.

Matches the changes done in vibe.d 0.8.x and vibe-core.

This still generates lots of deprecation warnings, but successfully compiles with vibe-core, as well as with the old vibe-d:core package.

If another package links against a vibe.d sub package instead of the base package, this now properly uses the vibe.d types instead of DUB's internal forks.

* Fixed output dir not being created on build complete.
* Fixed '/' being added to the front of Windows absolute paths.