Add documentation and example of arbitrary QAT flow used with recipe-based export.

[MartinPavella]

Summary

Adds a runnable example and documentation showing how to run an arbitrary Quantization-Aware Training flow in the middle of the declarative export recipe, by splitting the recipe around the QUANTIZE stage.

The declarative ExportRecipe / ExportSession path runs the whole pipeline as one export() call, which is fine for simple calibration but cannot host complex QAT — real training runs involve iterative loops, checkpointing, job restarts, validation-accuracy evaluation, rollback on regression, and possibly separate machines. The recipe's pipeline_stages attribute is used to run the pre-quantize stages, hand the graph off to an out-of-band QAT flow as a .pt2, and resume the recipe from TORCH_EXPORT onward.

1. New example under examples/export/qat_pipeline_split/: four independently-invokable stages driven by run.sh. 1_prepare.py captures up to the quantize boundary, 2_qat.py prepares with prepare_qat_pt2e and does a stand-in train + checkpoint round-trip before convert_pt2e, 3_lower.py lowers to model.pte via the recipe, and 4_run.py runs the .pte through the runtime. Each stage asserts its own invariants, so the example doubles as a smoke test.

2. The example shows one method with optional simplifications: the full split sets pipeline_stages explicitly on both halves, while the simplified path skips the pre-quantize slice and lets the recipe auto-skip SOURCE_TRANSFORM / QUANTIZE / TORCH_EXPORT for ExportedProgram input (default pipeline only).

3. New doc docs/source/using-complex-qat-flows-with-recipe-based-lowering.md, registered in the usage.md toctree and cross-linked from quantization-overview.md. using-executorch-export.md was left untouched since it documents only the imperative API.

Authored with Vero Code.

cc @mergennachin @nil-is-all @iseeyuan @lucylq @helunwencser @tarun292 @kimishpatel @jackzhxng

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22408

• 📄 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, 2 Unrelated Failures

As of commit 993330f with merge base a822ab2 ():

NEW FAILURES - The following jobs have failed:

• Lint / link-check / lint-urls (gh)
• pull / unittest / linux / linux-job (gh)
  pytorch_tokenizers
• pull / unittest-buck / linux / linux-job (gh)
  pytorch_tokenizers
• pull / unittest-editable / linux / linux-job (gh)
  pytorch_tokenizers

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• pull / test-arm-backend-no-driver (test_pytest_models_tosa) / linux-job (gh) (matched linux rule in flaky-rules.json)
  The runner has received a shutdown signal. This can happen when the runner service is stopped, or a manually started runner is canceled.
• pull / test-qnn-testsuite-linux / test-backend-linux (qnn, models) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

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

[linux-foundation-easycla]

• ❌ - login: @MartinPavella / name: Martin Pavella. The commit (993330f) is not authorized under a signed CLA. Please click here to be authorized. For further assistance with EasyCLA, please visit our EasyCLA portal and chat with our support bot.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.