Support alternate monthly mean calculations for tipping bucket stations

[corviday]

Stations with the network_variable_name of cum_pcpn_amt seem to measure precipitation by accumulation - that is, they report how much total precipitation has fallen since the last time the instrument reset on an hourly basis, instead of reporting how much precipitation has fallen that hour. Here are a couple representative graphs:

Physically it seems likely these are "tippng bucket" stations, measuring rainfall via a bucket that accumulates water until it is full, and tips out to reset.

In order to calculate monthly precipitation observed at these stations, we do not want to average hourly or daily reported values; this leads to months where, in the case of the second graph, every day or hour sees 1600 mm of precipitation! I believe the right calculation goes something like:

month_max = maximum value recorded during month
month_min = smallest value recorded during month
month_last = last value recorded during month
month_first = first value recorded during month

if month_max > month_last: #bucket "tipped" this month
   return (month_last - month_min) + (month_max - month_first)
else: bucket did not "tip" this month
    return month_last - month_first

In order for the back end to calculate this special case reasonably quickly, we need a new materialized view with month_max, month_min, month_last, and month_first for each station and month using the cum_pcpn_amt variable, as well as the earliest and latest days each month for which data is available so we can prorate it if there is less than a full month's data in the database.

[corviday]

Currently trying a different approach to this problem, see #216