Split hcAG_coarse into total and disaggregated data structures

[r-ash]

Closes #351

This will make it more consistent with inputs in the adult model.

Ignore the goals tests for now, fixing those will require an update to Spectrum reading code to supply new split data, I'll get that updated once this is in.

[r-ash]

@jeffeaton making this change to drop hcAG_coarse and use hcAG_c instead suggests me that this should be 2 here? But that was not in your original changes.

This is only used for hc_nosocomial_infections which is indexed using hc_age_coarse so it was fine. Updating this should just mean that drops an unused row from hc_nosocomial_infections when age10_adult is true.

[mwalte10]

I have a few suggestions of edge cases to check as i remember when coding this I would get the code to work in one pattern of ART entries and then it would instantly break.

Would be good to test that this survives if you:
go from 0-14 number entry to age disaggregate number entry
age disaggregate numer entry to 0-14% entry.
Those two patterns are what the vast majority of spectrum files will have.

[r-ash]

I have a few suggestions of edge cases to check as i remember when coding this I would get the code to work in one pattern of ART entries and then it would instantly break.

Would be good to test that this survives if you: go from 0-14 number entry to age disaggregate number entry age disaggregate numer entry to 0-14% entry. Those two patterns are what the vast majority of spectrum files will have.

Thanks Maggie, I tested this by creating a file with the following child ART input:

1. number -> percent (year 2022)
2. percent -> number (year 2017 and 2021)
3. number -> number age disaggregated (2018)
4. % -> number age disaggregated (2030)
5. number age disaggregated -> number (2031)
6. number age disaggregated -> % (2019)

And I could see this was breaking when t-1 was a non-age specific year as we we're only accumulating from the age disaggregated and not the total.

I've now fixed this and in comparing between this branch and current main the numbers agree up to floating point precision. So I think safe that this PR does not alter the calculations.

[mwalte10]

Looks good!