Change streams may be subject to spurious "CappedPositionLost" when resuming

Our testing infrastructure uncovered a rare case where this might happen, detailed in SERVER-49690. When a change stream is resuming, it may encounter this error. As far as I know this has never been observed, but I see no reason it couldn't happen. I would recommend looking into whether we can reproduce this. If so, I think we should do one of the following:
1) Disabling yielding when doing the oplog check upon resume during the change stream
2) Adding a similar retry loop within the change stream
3) Ensuring drivers will retry this error

During SERVER-49690 I looked into option #1 but the patch quickly exploded. I'll attach my WIP but it certainly won't compile and doesn't plumb the yield policy far enough to fix the issue.