Bug: Race-like situation leads to nil-pointer panic in replace cursor

[etiennedi]

Discovered during the investigation of #1849:

In the following place the Lock is obtained to late. A compaction can (and as our stress tests show will) happen between creating the slice and filling it. Thus we leave empty slots which will lead to a nil-pointer panic down the line

func (s *SegmentGroup) newCursors() ([]innerCursorReplace, func()) {
  out := make([]innerCursorReplace, len(s.segments))
  s.maintenanceLock.RLock()

  for i, segment := range s.segments {
    out[i] = segment.newCursor()
  }

  return out, s.maintenanceLock.RUnlock
}

Notes

• Check if the same issue also occurs for the other two cursor types (map/set) and fix if required.
• Validate either through direct test or through external stress-test

[etiennedi]

Update: Nothing to do for the other two types (map/set), there the lock is in the right place.