Answer a MEOS base value as the class the model gives it

[estebanzimanyi]

The object layer types a method by the class the catalog assigns each C type,
and the value types had none: a method taking a geometry, a text, a circular
buffer or a network point was deferred to the flat surface however ordinary it
was. The catalog carries a class for each of them now, so those methods reach
the object layer — Cbuffer.Point() answers a Geo, TGeo.AtGeom(Geo)
restricts a trip to a polygon, Text.Upper() answers a Text — and the layer
holds 108 classes and 1122 methods against 94 and 646, with 171 deferrals
against 246.

What a class's instances are a pointer to comes from the catalog as well. The
generator held a map of nine C structs, which is what made a class the model
gains one the binding cannot reach: it reads classes.<Class>.cType instead,
and wraps a pointer in the shallowest class carrying that struct, so a
GSERIALIZED * is a Geo with Geometry and Geography distinct classes
under it. The companion hierarchies are read from the model too, rather than
named, so a hierarchy it gains needs no edit here.

Four tests read the values back: a text upper-cased and title-cased, a
geometry's SRID and text, a circular buffer answering its centre as a geometry,
and a temporal geometry restricted by one, which ends where the trip leaves the
polygon rather than where the trip ends.

The deferrals left name no MEOS class: an Interval *, the uint8_t * WKB
buffers with their size_t * length, the double-pointer out-parameters, and
the vendored JsonPath *, PCSCHEMA * and Match * structs.