Optimize LazyUDF evaluation when using postfilters or pre filters

[martaiborra]

The blosc2.NDArray.__getitem__ method uses ndindex, which consumes a lot of time. This __getitem__ is used for each block and each blosc2.NDArray operand in a LazyUDF computed with prefilters or postfiters. It would be interesting to optimize the aux_udf function by calling the b2nd_get_slice_cbuffer C function instead of the __getitem__ method.