Treat aten.slice.Tensor as a shared-observer op in XNNPACKQuantizer

[Hyungkeun-Park-Nota]

Summary

_is_share_obs_or_fq_op lists aten.slice_copy.Tensor but not aten.slice.Tensor.
The quantizer runs before to_edge(), so the op it actually sees is
aten.slice.Tensor, the _copy entry never matches, and propagate_annotation
does not give the slice its producer's SharedQuantizationSpec.

The other data-movement ops in that list are registered under both names
(permute.default/permute_copy.default, squeeze.dim/squeeze_copy.dim,
view.default/view_copy.default); slice only had the _copy one. slice has
no .default overload, which is why the pre-functionalization name is
slice.Tensor and not slice_copy.default.

Without the shared spec, the slice input and output end up with different quant
params. XNNPACK's static_slice requires them to be equal, so the model
partitions and lowers fine and then fails at runtime init:

[XNNCompiler.cpp:1421] Failed to create static slice node 377 with code: xnn_status_invalid_parameter
[XNNPACKBackend.cpp:119] XNNCompiler::compileModel failed: 0x1
[method.cpp:110] Init failed for backend XnnpackBackend: 0x1

I hit this quantizing a Gemma-family embedding model with XNNPACKQuantizer
(per-channel W8A8 static), where all 96 slices from the RoPE rotate_half
pattern were left without quant params.

_test_qs8_slice_copy was already disabled for this ("T156004676 - slice isn't
propagated"), and its comment names the same cause, so I re-enabled it here.

Test plan

pytest backends/xnnpack/test/ops/test_slice_copy.py -k qs8

The test runs end to end now: quantize, lower, serialize, run and compare
outputs. Reverting the one-line change makes it fail with 2
quantize_per_tensor nodes instead of 5.

Two assertions in it needed updating first. check_node_count takes op
overloads rather than the "aten::slice.Tensor" strings the test was written
with, so it reported Expected 3 aten::slice.Tensor nodes but found 0 on a
graph that did contain them. And the expected quantize_per_tensor count is 5,
not 3, since the three slices are now quantized along with the input and the add
output.

lintrunner -m upstream/main is clean.

cc @GregoryComer @digantdesai @cbilgin @JakeStevens

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21568

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Github-hosted ubuntu-24.04 runners queueing

❌ 9 Pending, 1 Unclassified Failure

As of commit 208a8e2 with merge base 4b4df96 ():

UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:

• Check Labels (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)

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

[Hyungkeun-Park-Nota]

@pytorchbot label "release notes: xnnpack" "module: xnnpack"