Implement a Transformer Encoder only model
Map [past_covariates+old_target, future_covariates] to [old_target, target]
Alternatively, map [B_old, I_new] to [B_old, B_new]
Outcome
Implemented in transformertf : https://gitlab.cern.ch/dsb/hysteresis/transformertf/-/commit/a7145953929a7751ac1dff661d045d6459cde0d2