Append support for MultiPolygons in shapefiles

[olt]

Shapefiles have no type MultiPolygon, but they support them anyway (all rings are stored in one polygon). See: http://www.gdal.org/ogr/drv_shapefile.html

Fiona/OGR handles everything right when I create a new Shapefile with geometry type MultiPolygon, but when I open it for appending the type changes to Polygon. Fiona will then refuse to write additional MultiPolygons (see WritingSession.writerecs).

How should we handle that? Ignore the Multi prefix when checking the geometry type for 'ESRI Shapefile' drivers?

[olt]

Here is a possible, a bit hackish:
olt/Fiona@sgillies:master...olt:shapefile-multipolygons

I also improved the error output if the property schema differs.

[tylere]

I feel that this is an important issue. It is quite common to have shapefiles that contain multipolygons.

Here is a convoluted workaround for processing Shapefiles that contain MultiPolygon geometries.:

1. Read a multipolygon shapefile with Fiona
2. Iterate through the features
   * If it is a polygons geometry, use Shapely to convert the feature to a MultiPolygon
   * Process the MultiPolygon
   * Write the MultiPolygon to a GeoJSON output
3. Using ogr2ogr, convert the GeoJSON file back to a Shapefile

[sgillies]

(Info copied from #26)

See http://www.gdal.org/ogr/drv_shapefile.html:

Note that when reading a Shapefile of type SHPT_ARC, the corresponding layer will be reported as of type
wkbLineString, but depending on the number of parts of each geometry, the actual type of the geometry for each
feature can be either OGRLineString or OGRMultiLineString. The same applies for SHPT_POLYGON shapefiles,
reported as layers of type wkbPolygon, but depending on the number of parts of each geometry, the actual type
can be either OGRPolygon or OGRMultiPolygon.

When reading from a "polygon" type shapefile, you might get polygons or multi-polygon features. TBD: whether to coerce geometries or just relax the single-type constraint in Fiona for just the Shapefile format.

[sgillies]

Good grief, how did I overlook this issue!? @olt, @tylere, thanks. I've got a fix in commit 4632128.

Sometime I think it would be nice to do away with the distinctions between polygons/multipolygons and linestrings/multilinestrings for all data sources.