fix the replace slice pass in case the first dimension isn't 1.
This wasn't caught by CI because any sensible model has a batch size of 1, but caused an infinite loop when converting a pre-trained mobilenet_v1 (which included shape -> strided_slice ops later removed by the canonicalizer pass)
This wasn't caught by CI because any sensible model has a batch size of 1, but caused an infinite loop when converting a pre-trained
mobilenet_v1
(which included shape -> strided_slice ops later removed by the canonicalizer pass)