Speed up Store.last-modified by comparing hashes instead of file contents

[zazedd]

This commit changes the Store.last_modified function to compare the precomputed file hashes instead of their contents, as the latter is significantly (~10 to 20x) slower. :)
(More specifically, Tree.find is way slower than Tree.find_tree)

---

It also changes the documentation of the function where variable n is wrongly referred to as number.

[art-w]

Thanks, this provides a huge speedup for large values!

Looking at the overall algorithm used by last_modified, I believe there could be complexity explosion with fork/merge: it doesn't detect that a commit was already checked when reached from another path? I wonder if you have some ideas on how to fix this also :)

[clecat]

LGTM, thx a lot for the work
There might be a small issue however, when a commit is older than it's parent, the function might missbehave.