Feature/fix Labelselect folding and optimize output

1000 labels with low folding creates very large, long-running test cases

Feature/merge models

[Transform] Restructure MergeONNXModels to correctly preserve quantization information of non value_info tensors

Feature/mv chw add past conv

feature/refactor remove identity

Feature/move flatten past topk

Feature/move flatten past affine

hopefully speeds up GHA tests a bit

Feature/absorb transp in flatten

Feature/check int input before infer conv inp gen

Feature/mv transp past scal mul

Merge branch 'feature/check_int_input_before_infer_ConvInpGen' of https://github.com/quetric/finn into quetric-feature/check_int_input_before_infer_ConvInpGen

Feature/add downsampler op