The whole point of this effort is to replace the local file system
operation of my R scripts and allow multiple computers to do the same
thing, and to save the processing status in a CouchDB so that the
documents hold state of processing.

The documents hold state of processing.  Each vds id or wim site
number has various states that are either on or off, and that more or
less must be passed through in order.

1. They have to be loaded with raw data, per year.
    * The raw data has to be aggregated to hours.
    * The raw data has to be self-imputed to fill missing holes.

2. Each VDS site must be associated as a `neighbor` to at least one WIM
   site and direction so as to enable truck imputations.

3. Each unique WIM site and direction must be associated as a `pair`
   with at least one VDS site per year to enable the creation of the
   primary imputation set. The pairs must be drawn from the neighbors
   list.

4. Each VDS site for each year must be run through the truck
   imputation program to estimate the numbers of trucks at the site.

5. Each truck imputation output must be loaded into the database.


In addition to these states, there are states that exist for
geographic areas. Because these states depend upon the change in state
for member detectors, they can and should also be tracked in this
database.

1. Each geographic area uses the imputed truck and vehicle data to
produce aggregate estimates of volumes, speeds, etc.

2. Each time the imputations in the database are changed or updated or
added to, step 6 must be run again for each affected area.


Each of these states should be gettable, and most should also be
settable by external processes, so that the processing work can be
split among multiple servers.  Some, for example setting whether an
area needs to be recomputed, can be set by logical rules.  The ideal
tool would be a changes listener.

A flexible approach that moves logic out of the server would be to
just save all states in a single doc per detector or whatever, and any
external process gets the whole document.  This would mean the server
knows nothing about the docs it is shipping around, other than where
to find them.  That is good, as then the node server can be a low
level shell for CouchDB, and I can also just directly access CouchDB
from the R jobs if they are local to UCI.