Fix control flow output memory aliasing

[JacobSzwejbka]

Summary

Fixes #21490.

Control-flow branches intentionally share a memory arena, but their tensor outputs escape through MoveCall. The planner previously allowed a later higher-order operation to reuse branch storage while the earlier logical output was still live.

Memory planning now assigns each higher-order-op submodule arena a lifespan at its call site. For cond, that lifespan extends through the last use of its logical outputs. The true and false branches retain the same arena location because they are mutually exclusive, while consecutive conditionals receive non-overlapping storage when their outputs are simultaneously live. The emitter continues to use MoveCall; no tensor copy is added.

Regression coverage checks the original nested-condition and mutable-state failure, verifies non-overlapping allocations for consecutive conditional outputs, and asserts that branch results are emitted with MoveCall.

Test plan

pytest -q exir/tests/test_memory_planning.py::TestCond::test_consecutive_cond_output_lifetime exir/emit/test/test_emit.py::TestEmit::test_emit_cond exir/emit/test/test_emit.py::TestEmit::test_emit_cond_output_lifetime (3 passed)

pytest -q exir/tests/test_memory_planning.py -k "not custom_kv_cache_shared_buffers" (37 passed, 1 deselected; the installed custom-op library is binary-incompatible with the local Torch build)

pytest -q exir/emit/test/test_emit.py::TestEmit -k "not test_emit_sym_ite and not test_emit_sym_not" (68 passed, 2 skipped, 2 deselected; the installed ExecuTorch 1.3.1 runtime does not contain those current prim kernels)

python -m ufmt check exir/memory_planning.py exir/tests/test_memory_planning.py exir/emit/_emitter.py exir/emit/test/test_emit.py

python -m flake8 exir/memory_planning.py exir/tests/test_memory_planning.py exir/emit/_emitter.py exir/emit/test/test_emit.py

Authored with Codex.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21818

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit 58f7f6a with merge base 914705c ():
💚 Looks good so far! There are no failures yet. 💚

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

[JacobSzwejbka]

This approach is wrong gonna try to fix the memory planning issue rather then just deepcopying