Align function names and fix NAD sentinel for cross-platform SQL portability

[estebanzimanyi]

Background

MobilityDB Discussion #861 proposes a portable SQL dialect that runs the same query unchanged on MobilityDB/PostgreSQL, MobilityDuck/DuckDB, and MobilitySpark/Spark (see also MobilitySpark RFC #3).

MobilityPySpark has two gaps that block portability:

1 — Function name mismatches

MobilityPySpark uses Python-style names that differ from the MEOS/MobilityDB canonical names. The portable dialect requires canonical names so the same SQL runs on all three platforms:

MobilityPySpark (current)	Canonical (MEOS / MobilityDB)
tpoint_at(trip, instant)	atTime(trip, instant)
ever_intersects(trip, geom)	eIntersects(trip, geom)
min_distance(t1, t2) / nearest_approach_distance(t1, t2)	nearestApproachDistance(t1, t2)
at_period(trip, period)	atTime(trip, tstzspan)
at_geom(trip, geom)	atGeometry(trip, geom)

2 — NAD failure sentinel (blocks Q5 portability)

nearestApproachDistance / min_distance currently returns -1.0 on failure. The MEOS C layer uses DBL_MAX (established in MobilityDB PR #846); at the SQL level the canonical value is NULL.

Any UDF wrapper must convert the internal failure sentinel to None (→ SQL NULL) before returning. The -1.0 convention pre-dates PR #846 and should be retired.

Proposed fix

Rename UDFs to canonical names in pysparkmeos/UDF/udf.py and fix the NAD sentinel. Aliases for backward compatibility are optional but the canonical names must be the primary registered names.

This is the concrete change that unblocks BerlinMOD Q5 on MobilityPySpark.

cc @Action52