Before we build a potentially very big list of all the excluded file matching
every exclude pattern, which could be a lot and then do a containment check on
the resulting list.

A first step to speed up the process is to returns a set instead of a list so
containment checks will not be dependent to the number of matching files
anymore.

If the set still ends up being too large, introducing an intermediary solution
that check if a file name match a glob pattern instead of expanding the glob
pattern might be a good solution. Right now, it seems overkill in comparison
of this patch.

With this patch, building my project that exclude `nodes_modules/**/*` went
from 4 minutes to 7 seconds.

Fix #603

Fix a couple minor typos while we're at it

* Add a new test for the case where no vcs is detected

It can happens with unsupported VCS or no VCS at all (building from an
archive).
I copy the test_complete test and run it in a temporary directory where no VCS
should be detected.
I also modified the complete directory fixture directory, I can create a new
one if it's better

* Fix find_excluded_files to works without vcs

Fix #585

This change updates the method parameters used when calling
`PipInstaller.install_directory` to remove the `update` parameter.

This reverts commit dffb895f.

This is due to the fact the the python_full_version marker was not being used