Update on "Add preprocess to ci"

1. Run `test_preprocess.py` in ci.
2. Refactor `test_preprocess.py` by creating the model artifacts (eager, exported, et) once and re-use them on different inputs (previously, create model artifacts for each test).

Test Plan:
```
pytest -c /dev/null -v -n auto examples/models/llama3_2_vision/preprocess/
```

[ghstack-poisoned]

Author: lucylq

examples/models/llama3_2_vision/install_requirements.sh

@@ -10,4 +10,4 @@ pip install --pre torchtune --extra-index-url https://download.pytorch.org/whl/n
 
 # Install torchao.
 TORCHAO_VERSION=$(cat "$(dirname "$0")"/../../../.ci/docker/ci_commit_pins/torchao.txt)
-pip install --no-use-pep517 "git+https://github.com/pytorch/ao.git@${TORCHAO_VERSION}"
+pip install "git+https://github.com/pytorch/ao.git@${TORCHAO_VERSION}"