Be a good Vim citizen and don't load if `b:did_ftplugin` is already set.

Clean up the check for `g:no_plugin_maps` based on this style from
tpope's gitrebase.vim:

    https://github.com/tpope/vim-git/blob/5fcaf2b4f66cd28984cf4fe814d7803bbf073a12/ftplugin/gitrebase.vim

Allow the plugin's actions to be undone. Will be executed automatically
when users change the filetype, as explained here:

    :h undo_ftplugin

Instead of forcing the mapping '<C-x><C-s>' to activate Git SHA
completion, allow users to define a different custom mapping.

Don't force our functions to be loaded when the ftplugin is loaded. Be
good Vimscript citizens and respect load performance.

To ensure that we don't conflict with the real "gitcommit" ftplugin,
move ours into a new custom file.

As described in

    :h ftplugin-name

we can create additional ftplugins for the same filetype by putting our
script in a directory named for the filetype.

Make our custom mapping work while retaining any user-defined
`completefunc`.

Doing it this way enables us to get away without having to set a hacky
`CursorMovedI` unsetting autocommand as imagined in
e70d92c495001c0af0037b9fadbeec170d2c318f.

We clutter our `GitSHAComplete` completion function a bit with
functionality that doesn't really belong to it in order to get a win for
simple implementation of the `completefunc` restore feature.

Experiment with providing a custom mapping for our Git SHA completion
function. We use <C-x><C-s>.

The custom mapping should set our function as the custom `completefunc`,
call the user completion function by programmatically running
`<C-x><C-u>`, and then unset and restore the previous `completefunc`
value.

This allows us to not clobber a user-defined completion function and
also give ourselves a special custom <C-x> mapping.

Just experimenting for now. Figured out how to get this to work using a
test `ASDF` function along with these resources:
* http://stackoverflow.com/questions/15643234/remapping-tab-completions-in-vim
* https://github.com/ervandew/supertab/blob/master/plugin/supertab.vim
* http://stackoverflow.com/questions/6926034/creating-a-mapping-for-insert-mode-but-not-for-autocomplete-submode

It works. The only problem is that we can't restore the old
`completefunc` value. Thinking about ways to do this, and for now my
most promising idea is to use the `CursorMovedI` autocommand.

Get the list of SHAs from `git rev-list` and use this for
completion. Not too different from the original structure.

Renamed `res` to `matches` because that seemed clearer to me.