Add tensor_ext.partition op

[j2kun]

One of the ops for lowering key_switch_inner is to partition the coefficients of a polynomial. This op will eventually lower to a combination of

tensor.extract_slice (extract the divisible part)
tensor.reshape (reshape the divisible part)
tensor.extract_slice (extract the extra piece)

This actually kind of makes me think we don't need a dedicated op at all. We can just lower directly to the three ops this will lower to. Is there an advantage to having this intermediate op? I suppose that would depend on how other methods of implementing keyswitch might differ with this one.

Part of #2157

CC @crockea @AlexanderViand

[j2kun]

OK I added a second commit (so we can see them side by side) that shows what the lowering would look like if we skipped having a dedicated partition op and lowered to what tensor_ext.partition would have lowered to anyway. It seems tidy enough that I think we should just do that.