Improve precision for non-power-of-2 scales in brevitas -> QONNX workflow and add pytests

[JanFSchulte]

Adds simple pytests for the brevitas -> QONNX -> hls4ml workflow.

Tested are models using both power-of-2 and non-power-of-2 scales. For the latter, the accuracy is generally poor, driven by insufficient precision for the scale variable in the rescaling operation. So far I have only come up with a stupid fix that chooses an arbitrary higher precision to improve things, and I would like to discuss better options to infer the required precision.

Type of change

• Other (Specify)

Mostly adds tests

Tests

This PR adds the relevant pytests to test the brevitas -> QONNX -> hsl4ml workflow for a simple model with a linear layer and activation. More tests can be added, if desired.

Checklist

• I have read the guidelines for contributing.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation.
• My changes generate no new warnings.
• I have installed and run pre-commit on the files I edited or added.
• I have added tests that prove my fix is effective or that my feature works.