Fixed setting .next_if_* rules when multiple files are loaded: if
already set, do not change. This prevents loss of all exclude rules
after an inverted rule in a file and vice-versa.

When searching for .gitignore files in ancestor directories, ignore
empty path segments (typically at the beginning of an absolute path).

Directories were matched twice before: once when checking the parent
directory's entries and once when loading rules. Now both are done in
one step.

Git config files can contain duplicate sections (it seems that all but
one will be empty).

Two new arguments added to main():
  --zero -z: Use a null as separator
  --sep -s:  Set an explicit separator

The default separator is a carriage return.

The git loader generates the rule graph from the bottom up so that no
rules short-cut to a decision. (Which was a bug introduced while
attempting to prevent exceptions caused by dangling rules; wrong way of
fixing it, past self!)

If `Loader.load_from_dir()` raises SkipDirectory the directory is
pruned. This allows skipping based on forward looking (ie. check a
directory for a known file/subdirectory and prune based on it's
presence).

The git loader prunes based on the presence of a `.git` file or
directory anywhere but the project root which would indicate a
submodule.

Removes the GitGlobPattern class, uses standard GlobPattern class
instead.

All known ignore files should now be loaded by the git loader.
README.md has been updated to reflect all the known file locations
(including config file locations for the `core.excludesfile` value).

Fixes footnote formatting errors and a incorrect id-link.

* .gitignore - filters python compiled files
* list_files/ - python package which so far supports only git