Sharding invalid geometries

The 'shard' process halves a geometry (alternating horizontally & vertically) until each shard complexity is below a fixed threshold.
This ensures that the PIP performance is fast and linear.

This process is tested and works well, however, I have seen some geometries which fail to split due to topology errors.
It's pretty rare, it doesn't affect any WOF geometries and less than 1 in 10,000 OSM geometries.

I'll open up a separate issue related to repairing invalid geometries, but this issue is simply to handle the case where sharding fails due to a topology error.

In the case where a shard fails to half, we should:

• discard both of the two child halves (the ones created at level+1)
• do not delete the parent shard (the one at level)

I attempted to fix this in the past but for some reason IsValid() is returning 1 despite the shard geom being invalid.
It might be that spatialite@5 is better at handling this because it uses LIBRTTOPO instead of GEOS.

Start by tracing the shard process and the handling around IsValid(), then compare how spatialite@5, GEOS, and LIBRTTOPO report topology errors. Done means that when a shard fails to split, both level+1 child halves are discarded while the level parent shard remains.