fix: faster picking when no slices are made

[henryiii]

Helps with #644. Ideally, I think the order of picking and slicing should be flipped, which would probably be faster. Empty reduce should still be swapped for a copy, though, so this code code is still needed even in that case.

[henryiii]

Current benchmark info:

PYTHONPATH=build pytest tests/test_benchmark_category_axis.py --benchmark-enable

---------------------------------------------------------------------------------------------- benchmark 'Pick': 2 tests ----------------------------------------------------------------------------------------------
Name (time in us)               Min                     Max                    Mean                 StdDev                  Median                    IQR            Outliers         OPS            Rounds  Iterations
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_pick_only             469.6110 (1.0)       10,963.2400 (1.0)          548.6303 (1.0)         335.7542 (1.0)          496.3605 (1.0)          58.5930 (1.0)         7;153  1,822.7210 (1.0)        1054           1
test_pick_and_slice    752,036.0140 (>1000.0)  794,440.9860 (72.46)    768,945.9046 (>1000.0)  15,980.4104 (47.60)    764,982.2400 (>1000.0)  18,759.4093 (320.16)        2;0      1.3005 (0.00)          5           1
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------