Represent time using years rather than surveys

[athowes]

• More justifiable to have estimates for every year, especially with respect to the temporal random effects
• Infill "missing" years -- going to be a lot more rows
• Allows systematic approach to forwards forcasting of projections. But there is little temporal trend anyway
• Sometimes there are multiple surveys in the same year in a country e.g. UG2011AIS, UG2011DHS. How should the temporal random effects deal with this? Currently sur_idx = to_int(survey_id) (this likely isn't very robust). Have additional survey random effects

Subsumes #015

[athowes]

Note that the relevance of this increases with more surveys which might have overlapping years e.g. #41

[athowes]

See the bubble/dot plot as produced by #43 for evidence as to how much of a problem this might be.

[athowes]

• Study random effects
• Joint fitting of all countries #40 allows proper use of temporal random effects

[athowes]

Moving this task to #40.