Fix AOT non-power-of-two innermost sizes

[voltjia]

Summary

• Add an AOT regression test for static non-power-of-two innermost sizes.
• Convert positive integer constants used for innermost indices to the next power of two during code generation.
• Preserve the original logical size in masks, so generated tl.arange extents are valid for Triton while tensor semantics stay unchanged.

Testing

pytest output:

============================= test session starts ==============================
platform linux -- Python 3.10.16, pytest-9.0.2, pluggy-1.6.0
rootdir: /home/huangjiacheng/ninetoothed
configfile: pyproject.toml
plugins: anyio-4.12.1, xdist-3.8.0, cov-7.0.0, typeguard-4.4.4
collected 216 items

tests/test_add.py .                                                      [  0%]
tests/test_addmm.py ..                                                   [  1%]
tests/test_aot.py ............                                           [  6%]
tests/test_aot_auto_tuning.py ....                                       [  8%]
tests/test_attention.py ........                                         [ 12%]
tests/test_auto_tuner.py ....                                            [ 14%]
tests/test_clone.py ....                                                 [ 16%]
tests/test_conv2d.py ....                                                [ 18%]
tests/test_data_ptr.py .                                                 [ 18%]
tests/test_debugging.py .                                                [ 18%]
tests/test_dropout.py .                                                  [ 19%]
tests/test_eval.py ........                                              [ 23%]
tests/test_expand.py .                                                   [ 23%]
tests/test_generation.py ............................................... [ 45%]
.............................                                            [ 58%]
tests/test_getitem.py ..........                                         [ 63%]
tests/test_ipynb.py .                                                    [ 63%]
tests/test_jagged.py ................                                    [ 71%]
tests/test_matmul.py ..                                                  [ 72%]
tests/test_max_pool2d.py ..                                              [ 73%]
tests/test_naming.py .......                                             [ 76%]
tests/test_pad.py ................................................       [ 98%]
tests/test_pow.py .                                                      [ 99%]
tests/test_softmax.py .                                                  [ 99%]
tests/test_unsqueeze.py .                                                [100%]

======================= 216 passed in 952.02s (0:15:52) ========================