Answer NULL where the MEOS function has no answer to give

[estebanzimanyi]

A generated scalar binding returns SQL NULL where the MEOS function's
PostgreSQL wrapper does, as the catalog states it in shape.nullableResult: a
three-valued predicate answering below zero and a distance answering the
sentinel DBL_MAX. scalar_exec turns the stated guard into an ExecuteWithNulls
body marking the row invalid, as emit_geo_temporal marks its three-valued
predicates, and every by-value scalar emitter reads its executor, lambda and
return statement from it. A function whose catalog entry states no such guard
keeps the plain executor.

Witness. eDwithin(tgeompoint '[POINT(0 0)@2000-01-01, POINT(0 0)@2000-01-02)', tgeompoint '[POINT(0 0)@2000-01-02, POINT(0 0)@2000-01-03]', 300.0) answers
NULL, where (r != 0) read the -1 of two trips sharing no time as true, and
nearestApproachDistance of the same two trips answers NULL rather than
1.7976931348623157e+308. 064_tpoint_distance gains the two queries.

Measured. Regenerated against the catalog at the recorded MEOS commit
90009a793a, the surface holds the same 1870 bodies and 6680 registrations; 130
bodies change, each gaining its guard, and every one of the 207 emitted
functions whose catalog entry states if (result < 0) or
if (result == DBL_MAX) returns NULL on it. Over the 4781 candidate pairs of
one day of AIS trips in a 210 km2 belt, eDwithin at 300 m and a nearest
approach under 300 m disagreed on 218 pairs sharing no time, all read true by
eDwithin, and agree on every pair with the change. The 107 test files pass,
3053 assertions.

Why. A binding answering a value where MobilityDB answers NULL changes the
result of a query, and a benchmark written against the MobilityDB surface
counts 188 vessel pairs within 300 m where there are 182.