Remove unnecessary graph rewrites for both cuda and whole ET pipeline

[Gasoonjia]

CUDA / AOTI backend path

• Skip run_decompositions() during AOTI preprocessing when none of the backend decomposition table’s operators occur in the graph.
• Remove the obsolete CUDA int64 floor-div rewrite pass, since AOTInductor now handles the original operation directly.

Overall ExecuTorch export pipeline

• Only do operator decomposition when the operator exists in original edge graph.

Performance

MG 30B cold export time improved from 26:06 to 21:16.45, an 18.5% reduction.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21799

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 4 New Failures

As of commit f1f5ece with merge base e60faa2 ():

NEW FAILURES - The following jobs have failed:

• pull / android / run-emulator (gh)
  java.lang.IllegalStateException: Cannot access system provider: 'settings' before system providers are installed!
• pull / unittest-nxp-neutron / linux-job (gh)
  ../../dev/ir/converter/node_converter/test_hardtanh_converter.py::TestHardTanh::test__qat[PTQ-Inplace-activation_range1]
• Test CUDA Builds / export-model-cuda-artifact (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job (gh)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
• Test CUDA Builds / test-models-cuda / linux-job (gh)
  RuntimeError: Command docker exec -t e1910f7268be17a11d9844ff4bb8c3b81a619945263802c31e8b305ac1352a73 /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.