[Tests] Increase maximum quantization error

[kylesayrs]

Purpose

• Reduce false-positive test failures as a result of noisy quantization
  • https://github.com/vllm-project/llm-compressor/actions/runs/13775944340/job/38525122617#step:12:535

Changes

• Increase maximum error due to quantization from 0.022 to 0.025

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.