TODO

Create a house_no subtype to replace norm_addy.address, which we could then use to define 
a whole set of comparison operators on, since integers don't work everywhere.
--
Add the ability to limit the search to within a passed geometry.
--
if we are already loading a file, say, cousub from the state level,
we should not append the county level equivalent files.
---
It would be nice if we could find a good way to supress
shp2pgsql's need to spit out lines like:
  Shapefile type: Point
  Postgis type: POINT[2]
---
The following should be equivalent, can we match them?:

3244 Highway 116 N, Sebastopol, CA, 95472
3244 Gravenstein Highway N, Sebastopol, CA 95472

Why do we have different addys for:

select * from geocode('3244 Gravenstein Highway,  Sebastopol, CA 95472');
select * from normalize_address('3244 Gravenstein Highway, Sebastopol, CA 95472');
                       addy                       |                      geomout                       | rating 
--------------------------------------------------+----------------------------------------------------+--------
(3244,,Gravenstein,Hwy,N,,Sebastopol,CA,95472,t) | 0101000020AD100000E6AB07A4F5B65EC0AB11A69328384340 |      0

 address | predirabbrev | streetname  | streettypeabbrev | postdirabbrev | internal |  location  | stateabbrev |  zip  | parsed 
---------+--------------+-------------+------------------+---------------+----------+------------+-------------+-------+--------
    3244 |              | Gravenstein | Hwy              |               |          | Sebastopol | CA          | 95472 | t

---

Remove all deprecated tables and views.

---
Create a series of test addresses to verify against.
---
from normalize_address.sql:
  -- FIXME: in the case where we have just
  -- a street and zip (e.g. 123 Something, 12345), and
  -- there happens to be a place named 'Something' in that state,
  -- we will not match the street because we think it's a place. 
  -- I'm not sure a good way to work
  -- around this.
  AND
  -- FIXME: we should have a lookup table like
  -- direction and street_types appropriate to the suffixes
  -- sufqual as prequal above.

--

address_variants() should be more thoughtful about ranking
an address based on the number of (mis-)matched elements.