Add lwtnn for pixel clustering NNs
This is for adding lwtnn-based NN reading. Validation of individual clusters has ensured that:
- The values returned by the MDN are as intended based on the training
- Estimated positions are very similar to those from the networks in production to be replaced
- Estimated errors are also quite similar, if slightly larger on average.
This is set up without any user-defined flag controlling whether or not the MDN is used. Instead, if the table is available in the COOL tag and the run being reconstructed is within the interval of validity, MDNs will be used.
On our side, we still need to update the contents of the COOL table to reflect a recent update to the offset values in the NN configurations, but this shouldn't hold up the MR. We should be ready to proceed.
Tagging @gfacini, @battagl , @goblirsc , @npetters for info.