[dynamo][fx] Don't emit call_function node to construct dataclass instances for Dynamo and make_fx tracing

[StrongerXi]

Stack from ghstack (oldest at bottom):

• [dynamo] Use the new get_unique_name_wrt helper when applicable #146950
• [dynamo] Initial support for nonstrict_trace #146367
• [hop] Support more output types for flat_apply #146714
• [dynamo][fx] Support dataclass whose fields have init=False #146713
• -> [dynamo][fx] Don't emit call_function node to construct dataclass instances for Dynamo and make_fx tracing #147152
• [dynamo][fx] Don't emit call_function node to construct NamedTuple instances for Dynamo and make_fx tracing #147145

As title. The behavior change is limited to Dynamo and make_fx tracing
for backward compatibility reasons with symbolic_trace.

It heps enforce the invariant that Dynamo and make_fx graphs would
always contain tensor ops -- so rather than having these call_function
nodes to construct NamedTuple, we inline them directly as instance
arguments to the user nodes.

cc @ezyang @SherlockNoMad @EikanWang @jgong5 @wenzhe-nrv @voznesenskym @penguinwu @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @jiayisunx @chenyang78 @kadeng @chauhang @amjames

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/147152

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

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

❌ 3 New Failures, 1 Unrelated Failure

As of commit 607542e with merge base f95bdf5 ():

NEW FAILURES - The following jobs have failed:

• Lint / lintrunner-noclang / linux-job (gh)
  >>> Lint for torch/fx/node.py:
• pull / cuda12.4-py3.10-gcc9-sm75 / test (pr_time_benchmarks, 1, 1, linux.g4dn.metal.nvidia.gpu) (gh)
  REGRESSION: benchmark ('aotdispatcher_training_subclass_cpu', 'compile_time_instruction_count') failed, actual result 10558710218 is 1.62% higher than expected 10390000000 ±+1.50% if this is an expected regression, please update the expected results.
• pull / linux-focal-cuda12.4-py3.10-gcc9-sm89 / test (default, 3, 5, lf.linux.g6.4xlarge.experimental.nvidia.gpu) (gh)
  dynamo/test_higher_order_ops.py::TestHigherOrderOpsOpInfoCUDA::test_hops_compile_backend_inductor_invoke_quant_simple_cuda_float32

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• inductor / unit-test / linux-jammy-cpu-py3.9-gcc11-inductor / test (inductor_avx2, 2, 2, linux.10xlarge.avx2) (gh) (trunk failure)
  'Test'

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

[StrongerXi]

So we have a real failure in inductor, since it was assuming a dataclass input to torch.ops.higher_order.invoke_quant_packed to be a node that constructs the instance, yet this PR changes it into the dataclass instance itself. @zou3519

pytorch/torch/_inductor/fx_passes/joint_graph.py

Lines 461 to 463 in edfe81e

	quant_options_node = kwargs.pop("quant_options", None)
	if quant_options_node is not None:
	assert isinstance(quant_options_node, torch.fx.Node)

[StrongerXi]

Abandoning this stack, see #147530.