Prediction models: Remove detection & use of target signal frequency in the Prediction Model (ML) and Classification Model pipelines. This was causing some model run failures, and is unnecessary as both model pipelines are agnostic to the frequency of data (they do not do any temporal aggregations).
DSL: Return NaN and not 0 when summing only NaN data points in .group_by_entity(). This fixes a bug with additional zero-value data points being introduced incorrectly in some cases.