Skip passing normalization if it is not active in the config

[dtronmans]

Purpose

Resulting ONNX config:

Resulting RVC4 config:

None for mean and scale is properly handled in modelconverter:

• https://github.com/luxonis/modelconverter/blob/main/modelconverter/utils/onnx_tools.py#L139
• https://github.com/luxonis/modelconverter/blob/main/modelconverter/utils/onnx_tools.py#L241
• https://github.com/luxonis/modelconverter/blob/main/modelconverter/utils/onnx_tools.py#L212

Another option would be to manually pass the identity (so [0, 0, 0] [1, 1, 1] for RGB and [0] [1] for grayscale) to the ONNX archive if normalization is not active, instead of passing None and letting it be handled downstream by modelconverter

Specification

Dependencies & Potential Impact

Deployment Plan

Testing & Validation

[klemen1999]

LGTM. And yes, I think is fine to have null values and let other tools (modelconverter) handle this correctly since we do allow None values for this parameters in the schema.