TreeDirNode.openAlong for Issues #179.

Add 2 autocmds - NERDTreeInit and NERDTreeNewRoot. These are called when
a tree is created and when the root is changed. The goal is to give the
fugitive plugin something to listen for so it can add the G* commands to
nerdtree buffers in git repo dirs.

When 'virtualedit' is set, the column where the cursor resides can no longer be
relied on for determining the number of elements in the 'line' list for
iterating.  Replacing virtcol() with len() seems to work correctly and not
result in E684 errors.  Fixes Issue #144

Previously if we used the gi/gs mappings on a dir node/bookmark the
cursor would end up in the new window. Now it stays in the current
window as expected

Previous case could fail with something like '<Tab><Tab>'.  Cases like
that should be handled correctly now.

KeyMap.bind() does not gracefully handle use of '<>' notation.  For
example, trying to call NERDTreeAddKeyMap() with a 'key' argument of
'<Leader>e'.  There were some workarounds KeyMap.bind() to help with
this by specifically allowing you to leave off the '<>' parts for
'<C-...>', '<M-...>' and mouse mappings and it would add them back for
you before creating the mapping.  This commit reverts some of that logic
and simply says that if the key starts with '<', replace it with <lt>.

These calls to NERDTreeAddKeyMap use the same callback as when these
keys are mapped with a 'scope' of "Node."  This should not pose a
problem though as s:previewNodeCurrent, s:previewNodeHSplit and
s:previewNodeVSplit all just call the passed in node's open() method,
which gets passed on to an Opener object.

this was broken in one of the recent "epic refactor" commits

The core issue is that in some versions of vim resolve() will remove trailing
slashes, while in others it will not.  This lead to commit
bc745b6e99 attempting to address a double slash
problem.  However, that broke symlink detection on systems where resolve()
removes trailing slashes.  This new function just calls vim's resolve()
function, but removes trailing slashes if they exist.

This reverts commit bc745b6e99.  This
broke symlink detection on systems where vim's resolve() removes the
trailing slash from a path.

This was broken at some point - fix it.

Dont do `remove(array_of_prototype_obs, index(...)))`. In the past this
has been found to cause seg faults when the objects get large.

Previously we are doing this every time a nerdtree window was created -
which is wrong and was removing custom mappings from users that overrode
default key mappings.

Now we only generate the defaults once.

Older versions of vim require you to unlet a dictionary function if you
are replacing it (i.e. overriding it in a prototype OO scenario). The
unlet call got lost in the recent refactoring.

Move the code to split windows and open nodes and bookmarks out into a
dedicated class. This will remove duplication and remove and centralise
a concern from the other classes.

this is to prevent autocmds from being fired when the tree is closed

Make these all go through .open() and pass in args to control the
behaviour. Deprecate the old `openSplit`, `openVSplit` and
`openInNewTab` methods and make them private.

This makes the API a lot cleaner - it was getting pretty messy in this
regard.

This allows users to specify whether each regex in NERDTreeIgnore should
apply to only files or only dirs.

This fixes #89.

First, set the tree parts regex appropriately depending on
g:NERDTreeDirArrows.

Second, fix an edge case when trying to edit a file like +foo where the
+ meant 'foo' was being interpreted as an arg to the :edit cmd

this gives users a chance to filter out nodes using whatever logic they
please

Just call these functions directly at the bottom of the nerdtree script.
This simplifies things and fixes #99.

It seems that some versions of vim done like the syntax of:
    function("foo")()

so refactor it slightly.