Another epically botched polling sample, as evidenced by the partisan breakdown
Again, you're misunderstanding how polling works. They sampled a large number of Republicans, because they want a large sample for the presumably competitive GOP primary race (whereas the Dem. race is less competitive, so they don't care as much about sample size). But the large number of Republicans in the sample are then weighted down to something more realistic when they compute the GE numbers. That's why they say:
"The weighting of the results was performed separately depending on whether the question was for all respondents, just Republicans, or just Democrats."
Now, that doesn't mean the poll is any good. Gravis isn't very good, based on their track record. But not based on the sample they used in this particular poll, since we don't actually know what partisan breakdown they're assuming in this poll.