Extension of Regression/Benchmarking CI

[fpjentzsch]

Current state is documented here: https://github.com/eki-project/finn-plus/wiki/CI-Test-Benchmarking-Suite

Enable Flow verification for all models

This requires a single golden truth I/O pair ("inp.npy" & "out.npy") that is either placed in the models dir (created once at export time) or generated from the input .onnx model at the beginning of the build. Ideally, all verification steps should be enabled.

• Transformers
• VGG-10
• MobileNetV1
• Cybersecurity MLP
• bnn-pynq tfc
• bnn-pynq cnv
• GTSRB
• KWS

Enable on-board validation for all models

• PR for enabling simultaneous performance profiling and inference on real data: Update Pynq Driver #100
• Fix and merge Faster data unpacking Xilinx/finn#1291 to avoid extreme packing overhead for ImageNet2012 dataset (and potentially others)
• Add driver support for validating on the following datasets
  • RadioML 2018
  • UNSW-NB15
  • GTSRB
  • Google Speech Commands v2
  • TinyStories
• Support deployment on Alveo in the CI context

Models to add

• Generic ResNet flow (e.g., ResNet-18, ResNet-8, ...) to replace RN-50 example
• MobileNetV2
• YoloV8
• ...

Micro-benchmarks to add

• MVAU
• VVAU
• SWG
• Thresholding
• Elementwise

General Improvements

• Parse/collect more reports
  • post_route_timing.rpt
  • OOC detailed resource report
  • Generate layer type breakdowns for all utilization reports
• Switch from GitLab pipeline variables to inputs where applicable
• Make logging to stdout and logfile more robust
• Allow tool version selection via input/variable

[fpjentzsch]

Moved documentation to CI Test Benchmarking Suite & Example Models, so this issue is used only for tracking planned improvements/extensions.