Hg storage now based off mercurial module, not 'hg' executible.

This should make repeated calls to Hg storage instances _much_ faster,
since we avoid repeatedly loading and tearing down a python subprocess.
For example, the testsuite runs ~6x faster on my box.

Here's a run with the old Hg implementation:
  $ python test.py libbe.storage.vcs.hg
  ...
  =================================
  ERROR: test_get_previous_children
  ---------------------------------
  Traceback (most recent call last):
    ...
  NotImplementedError
  
  ---------------------------------
  Ran 49 tests in 133.285s
  
  FAILED (errors=1)

A run with the new implementation gives the same results, except for:
  Ran 49 tests in 22.328s

1 files changed, 1 insertions, 1 deletions

diff --git a/libbe/storage/vcs/bzr.py b/libbe/storage/vcs/bzr.py
index 397267a..7335861 100644
--- a/libbe/storage/vcs/bzr.py
+++ b/libbe/storage/vcs/bzr.py
@@ -51,7 +51,7 @@ def new():
 
 class Bzr(base.VCS):
     name = 'bzr'
-    client = None # bzrlib
+    client = None # bzrlib module
 
     def __init__(self, *args, **kwargs):
         base.VCS.__init__(self, *args, **kwargs)