Use numba code for supported CAReduce cases

[ricardoV94]

Description

Apparently both out implementation and numba's can be pretty dumb for reductions. This PR tries to default to the Numba case when possible. I am not sure this is actually better, see the comparisons below, some cases improve by a lot but others get worse. Seems to depend on which axis are actually being reduced:

Summary:

• careduce_benchmark[Max-None] is 4x faster
• careduce_benchmark[Sum-0] is 100x faster
• careduce_benchmark[Sum-2] is 6.5x SLOWER
• logsumexp_benchmark[0-size2] is 1.3x faster
• logsumexp_benchmark[1-size2] is 1.6x SLOWER

----------------------------------------------------- benchmark: 40 tests -----------------------------------------------------
Name (time in ms)                                 Min                   Max                  Mean              StdDev          
-------------------------------------------------------------------------------------------------------------------------------
careduce_benchmark[Any-0] (before)            74.6352               86.0826               78.2833              3.1784          
careduce_benchmark[Any-0] (after)             72.2949               88.1325               74.8778              3.9049          
careduce_benchmark[Any-2] (before)            19.8382               22.8631               20.4905              0.8476          
careduce_benchmark[Any-2] (after)             19.8045               40.1451               22.8281              3.5232          
careduce_benchmark[Any-None] (before)         47.1671               49.2407               48.1692              0.7547          
careduce_benchmark[Any-None] (after)          47.0192               64.6291               51.9837              5.0230          
careduce_benchmark[Any-axis2] (before)        45.6168               49.6917               46.6328              1.2423          
careduce_benchmark[Any-axis2] (after)         45.3280               54.8637               48.6796              2.1270          
careduce_benchmark[Max-0] (before)           823.1772            1,072.3205              918.1883            121.3394          
careduce_benchmark[Max-0] (after)            895.3045            1,069.0787              950.5921             72.8457          
careduce_benchmark[Max-2] (before)            18.3019               28.5969               20.2784              2.2818          
careduce_benchmark[Max-2] (after)             17.7778               19.0539               18.2724              0.4013          
careduce_benchmark[Max-None] (before)         18.1158               19.4919               18.6922              0.4872          
careduce_benchmark[Max-None] (after)           5.3857                7.3592                5.7938              0.4722          
careduce_benchmark[Max-axis2] (before)        18.1986               20.8342               18.8722              0.6357          
careduce_benchmark[Max-axis2] (after)         17.9239               20.6664               18.9036              0.4819          
careduce_benchmark[Sum-0] (before)           939.0658            1,121.5008            1,002.5721             77.1697          
careduce_benchmark[Sum-0] (after)              8.7608               11.6043                9.8752              0.4344          
careduce_benchmark[Sum-2] (before)            14.9317               17.0656               15.6297              0.5323          
careduce_benchmark[Sum-2] (after)             92.5869              110.1285               96.9401              5.1967          
careduce_benchmark[Sum-None] (before)          6.0364                8.7782                6.4707              0.5112          
careduce_benchmark[Sum-None] (after)           5.2403                7.5870                5.9621              0.5091          
careduce_benchmark[Sum-axis2] (before)        14.8038               19.7759               15.8386              1.1597          
careduce_benchmark[Sum-axis2] (after)         14.7872               56.1588               19.9778              7.8031          
elemwise_speed (before)                        0.6115                1.9634                0.8762              0.1988          
elemwise_speed (after)                         0.5949                0.9883                0.6228              0.0600          
fused_elemwise_benchmark (before)              0.0657                0.2225                0.0731              0.0109          
fused_elemwise_benchmark (after)               0.0567                0.3472                0.0875              0.0348          
logsumexp_benchmark[0-size0] (before)          0.0143                0.0665                0.0156              0.0019          
logsumexp_benchmark[0-size0] (after)           0.0141                0.0460                0.0153              0.0018          
logsumexp_benchmark[0-size1] (before)         10.0825               11.5155               10.3202              0.2845          
logsumexp_benchmark[0-size1] (after)           9.5071               12.6158                9.8153              0.5590          
logsumexp_benchmark[0-size2] (before)      2,644.3215            3,462.0411            3,016.7679            296.5158          
logsumexp_benchmark[0-size2] (after)       1,899.5727            3,078.1175            2,229.7692            480.8884          
logsumexp_benchmark[1-size0] (before)          0.0146                0.0863                0.0172              0.0038          
logsumexp_benchmark[1-size0] (after)           0.0145                0.0567                0.0154              0.0015          
logsumexp_benchmark[1-size1] (before)          9.6068               13.1668                9.7613              0.3543          
logsumexp_benchmark[1-size1] (after)           9.5730               21.5240               10.2803              1.7398          
logsumexp_benchmark[1-size2] (before)      1,130.4573            1,481.2448            1,277.5037            164.1396          
logsumexp_benchmark[1-size2] (after)       1,848.0962            2,219.1350            1,986.3702            145.0611          
-------------------------------------------------------------------------------------------------------------------------------

Related Issue

• Closes #
• Related to #

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):