For buds with two mothers, we discard the second predicted mother.

validate_lineage returns the corrected lineage, with any extra
mother-bud assignments discarded.

Its checks for duplicated mothers became redundant and were deleted.

Relates to #52

WHY IS THIS CHANGE NEEDED?:
- IndexError raised when NameGrouper used to define a grouper object
  and then when .concat_signal() was invoked

HOW DOES THE CHANGE SOLVE THE PROBLEM?:
- effectively reverted to aliby@252915c6
- the two commits leading up to that one solved the problem: first by
  changing chainer's assumption that all experiments have a GFP channel,
  then updated references to renamed variables

WHAT SIDE EFFECTS DOES THIS CHANGE HAVE?:
- risks: may have broken somewhere upstream not otherwise specified yet

EVIDENCE THAT COMMIT WORKS:
- running grouper works

REFERENCES:
- issue #35