Try to fix random failures in test_recommend functional test
It's tricky because test itself contains randomization (perhaps more than it should).
The main problem is the test line:
assert_valid_recommendations(recs, max_articles=30, min_articles=15,
since_id=since_id)
but if we don't generated enough random articles such that there are at least 15 articles with article_id greater than since_id, then the test can fail, so make the number of random articles sent to the recsystem larger (which may also be useful for testing less trivial recsystems)