No capability for hindcast mode or variable future timesteps

[pkyle]

This is mostly a pointer to an issue documented on the PNNL Jira, briefly summarized here for those outside the firewall. The joining of "model_future_years" and "future_years" (intended to be future years for purposes of processing data from other sources) into a single constant ("FUTURE_YEARS") has had the unintended effect of disabling hindcast mode. Also, a few changes were needed both in gcamdata and in the old data system in order to enable the capacity to run with variable future timesteps. These issues are addressed in Core model proposal 276.