Use host system OSTree repo as local cache

When running the image builder on Endless OS, there is a high chance
that the host system will be running a similar, if not identical, OSTree
commit to the one being built into the image. In my case, I run master
and update essentially every morning, so running the image builder with
its default arguments will use the same OSTree as is booted.

`ostree pull` supports a `--localcache-repo PATH` argument:

> Like git's clone --reference. Reuse the provided OSTree repo as a
> local object cache when doing HTTP fetches.

On my developer system & internet connection (80 Mbps according to
fast.com), in the case where my system repo has the latest commit but
the eos-image-builder cache repo does not, the pull takes 15 seconds
with this patch. Without it, the pull takes, let's say, substantially
longer. I gave up after 10 minutes, after which time 26% of objects have
been pulled, topping out at around 900 kB/s.

run-build

@@ -95,11 +95,23 @@ class ImageBuildRoot(object):
         repo_path = config['ostree']['repodir']
         remote = config['ostree']['remote']
         ref = config['ostree']['ref']
+
+        # If we are on an OSTree system, use the system repo as an additional cache.
+        # In the case of an Endless OS developer system booted into the same or similar
+        # ref as we are building an image of, this saves pulling over the network. In
+        # other cases it is harmless.
+        system_ostree_repo = '/ostree/repo'
+        if False and os.path.isdir(system_ostree_repo):
+            extra_pull_args = ['--localcache-repo', system_ostree_repo]
+        else:
+            extra_pull_args = []
+
         log.info(f'Pulling OSTree ref {remote}:{ref}')
         eib.retry(subprocess.check_call, [
             'ostree',
             f'--repo={repo_path}',
             'pull',
+            *extra_pull_args,
             remote,
             ref,
         ])