Handle changed behavior in choose_multiple_weighted from rand 0.9

In rand 0.9, choose_multiple_weighted now returns an error if fewer than n items have a nonzero weight.

This is the likely cause of #1902 (closed); if so, !2877 (merged) should fix it.

We should:

• Look for every other place we call choose_multiple and choose_multiple_weighted and make sure we're handling this error case.
• Stop suppressing errors from these functions (at least we should log them!)
• Add a test for this behavior.