hgserve: don't scan for repositories
While investigated #67 (closed), it turned out that the cause for performance problems was
- the scanning of repositories by hgserve is way too slow
- it happens too frequently (by default 20s)
On the other hand, in #64 (closed), the problem is precisely that hgserve is not aware of a new repository until it has performed said scan…
Seen from a distance, this scanning is really useful to display the expected web pages to browse the repositories, something that we don't use at all in Heptapod.
It would be better to remove this scanning altogether, in favour of an on-demand approach, backed by some kind of LRU cache and some invalidation tricks for exceptional cases (ex: avoid funny things if a repo is deleted, and a new one is created at the same path)