[Bug Fix] Fix Aggregator For HAWQ Sensitivity Metric

[anzr299]

Changes

Change the formula for HAWQ aggregator

Reason for changes

Possibly not the right method.

Related tickets

186588

Tests

Ratio: 0.5; INT4_ASYM; GS128

Model	Variant	Strict	Flexible	Size
Qwen3-8B	buggy	84.91%	85.22%	6.665 GB
Qwen3-8B	fixed	85.82%	86.28%	6.663 GB
Qwen3-0.6B	buggy	29.57%	29.64%	0.500 GB
Qwen3-0.6B	fixed	31.84%	31.84%	0.500 GB
Llama-3.1-8B	buggy	72.10%	73.09%	6.503 GB
Llama-3.1-8B	fixed	72.63%	73.46%	6.503 GB
Llama-3.2-3B	buggy	57.62%	63.61%	2.594 GB
Llama-3.2-3B	fixed	57.62%	63.61%	2.594 GB
Gemma4-E4B	buggy	61.79%	60.96%	6.650 GB
Gemma4-E4B	fixed	61.94%	61.18%	6.650 GB

[Copilot]

Pull request overview

This PR updates the HAWQ sensitivity metric aggregator in NNCF to use a corrected accumulation formula, and adjusts the unit test reference values accordingly.

Changes:

• Fix HAWQ aggregation to accumulate a per-tensor normalized squared-sum instead of repeatedly renormalizing the running container.
• Update unit test reference outputs for the HAWQ aggregator to match the new formula.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/nncf/common/tensor_statistics/collectors.py	Adjusts HAWQAggregator accumulation logic (but needs a reset() override to avoid breaking reuse after TensorCollector.reset()).
tests/common/test_reducers_and_aggregators.py	Updates expected HAWQ aggregator outputs (and should add reset/reuse coverage).