Graceful service degradation

[isoos]

We don't cache the results from search service, but we could introduce a network-first caching strategy for it:

• we query the service the same way as now
• we store the result in memcache (TTL = 1 hour)
• when the service is down, we fall back to the cached results

This would not have prevented #1098 because its original cause was a memcache issue, but could be helpful in other cases, for short outages.

[jakobr-google]

We do need some kind of graceful degradation when search is down. At least show the static pages, and maybe a little "pub isn't feeling well" box where the top packages should be, if we can't reach search and memcache.

[isoos]

After the handler refactoring (with all the wrappers), if e.g. search is down and SearchClient throws an exception, we'd get something like this:

@jakobr-google: I think some search caching (e.g. packages for the landing pages) would be useful, but it can go on with a different tracking issue and priority. wdyt?