Updates since last week

• tpc-native-writer works only for small simulation, large simulations fail with
  [ERROR] websocket_server_callback: Error while reading from websocketconnection reset by peer
  [ERROR] Workflow crashed - PID 1961799 (tpc-qa-ideal) was killed abnormally with Segmentation fault and exited code was set to 139.
• Created independent simulation of 50Ev @ 50kHz PbPb and tested network vs. native performance. Digit maxima (raw): 26.0 mio., Ideal maxima (raw): 23.6 mio., Ideal clusters (after tagger): 19.2 mio.

• Interstingly CNN (top) does seem to do a better job for finding the loopers than the fully connected net (middle) (checked by example) and much better than the native clusterizer (bottom)

• On small simulation, clusterQC works -> Native file is written correctly, just websocketconnection holding it back for "big" simulation

• Managed to attach argmax layer to neural network with N class classification in ONNX -> Can run in current task, will start tests later