Remove SSM start_year parameter, replace with ref_year

[virgesmith]

This should be replaced by a census reference year (2001 or 2011), the SSM should also be able to work backwards, i.e. sequentially microsynthesise a 2001 population starting with 2011 census data and working backwards though the mid-year estimates 2010-2001.
This enables comparisons of 2011-based 2001 SSM against 2001 census as well as 2001-based 2011 SSM against 2011 census

[virgesmith]

Also remove projection_ref_year from json cfg - it's determined by the SNPP data and the parameter was never used