Lower memory usage for pin geometry calc

[wrridgeway]

We go from 1,469,467 to 1,470,359 rows (a difference of 892). Here is what the new data looks like compared to the old data, minus the geometry column:

skim_type	skim_variable	source	n_missing	complete_rate	charactermin	charactermax	characterempty	charactern_unique	characterwhitespace	numericmean	numericsd	numericp0	numericp25	numericp50	numericp75	numericp100	numerichist
character	pin10	new	0	1	10	10	0	1457959	0
character	pin10	old	0	1	10	10	0	1457163	0
numeric	start_year	new	0	1						2006.310948	1.852083596	2006.0	2006.0	2006.0	2006.0	2024.0	▇▁▁▁▁
numeric	start_year	old	0	1						2006.30021	1.800622516	2006.0	2006.0	2006.0	2006.0	2023.0	▇▁▁▁▁
numeric	end_year	new	0	1						2022.955853	2.291021408	2006.0	2023.0	2023.0	2024.0	2024.0	▁▁▁▁▇
numeric	end_year	old	0	1						2022.623576	2.186314394	2006.0	2023.0	2023.0	2023.0	2023.0	▁▁▁▁▇
numeric	longitude	new	0	1						-87.76344617	0.140398164	-88.2635061	-87.82816859	-87.73445373	-87.66719229	-87.5245681	▁▁▃▇▅
numeric	longitude	old	0	1						-87.76329732	0.140232267	-88.2635061	-87.82795164	-87.73439089	-87.66715148	-87.5245681	▁▁▃▇▅
numeric	latitude	new	0	1						41.84223347	0.167653113	41.46975644	41.71743355	41.85172987	41.9808197	42.15406191	▃▆▇▇▆
numeric	latitude	old	0	1						41.84211685	0.167632348	41.46975644	41.71734915	41.85154599	41.98052128	42.15406191	▃▆▇▇▆

[wrridgeway]

It's annoying to loop through NA values. I tried using both group_map() and split() to avoid this, but they seemed to take far longer than just walking through the possible values and using filter().

[wrridgeway]

This isn't necessary, but collecting the data and filling it takes a lot of time/memory and this let's us avoid having to redo it if anything goes wrong.

[wrridgeway]

We're just moving this work inside of a map function to let us load less data into memory. I've chunked everything by town_code since we only need to look back in time within pin10, not across pin10.