docs: parity-status snapshot + regeneration script

[estebanzimanyi]

A planning artefact, not a code change. Adds:

• `scripts/parity-audit.py` — walks MobilityDB's `CREATE FUNCTION` surface and MobilityDuck's `RegisterFunction` calls, writes a section-by-section coverage table.
• `docs/parity-status.md` — the current snapshot.

Match is by function name only, case-insensitive. A name registered in MobilityDuck is treated as covering all its overloads, so per-overload signature parity (e.g. `before(tstzspan, temporal)` vs `before(temporal, temporal)`) is not verifiable at this granularity.

Top-line numbers (snapshot 2026-04-30)

• 104 MobilityDB SQL section files, 2054 function declarations across them, 1032 unique names.
• MobilityDuck registers 359 unique names, 1653 total registrations.
• Name-level coverage: 1017 / 2054 (49.5%).

Where the gaps cluster

Bucket	Sections at 0%	Notes
Aggregates	`temporal/040_temporal_aggfuncs`, `temporal/042_temporal_waggfuncs`, `temporal/015_span_aggfuncs`, all `*_aggfuncs` per type	Aggregate framework partially being landed in PR #21. Each section has 7-40 transition / final / combine functions.
Index support	All `_gist`, `_spgist`, `*_indexes` sections	Internal helpers — never user-facing SQL. Will likely stay 0% until DuckDB grows comparable index extension hooks.
Comp / pos / topo operators	`_compops`, `_posops`, most `*_topops`	DuckDB rejects multi-char operator tokens; named-function equivalents land per-PR (`before`, `overbefore`, etc.). PR #64 catalogues the structural divergence in `docs/DuckDB-Parity-Gaps.md`.
Tile / split	`temporal/025_temporal_tile`, `geo/058_t{geo,point}_tile`	Manifest test landed in PR #64. Implementation pending.
npoint base type	`npoint/081_npoint` (15%), `npoint/091_*_routeops`	Base type only partially registered. Many npoint-network functions still unbound.
Analytics	`temporal/046_temporal_analytics`, much of `geo/076_*_analytics`	DouglasPeucker / simplify / minDistSimplify family. Easy adds; small section.

Sections at 100% coverage

`temporal/002_set_ops`, `temporal/005_span_ops`, `geo/072_tpoint_tempspatialrels`.

Sections close to done (>=80%)

`temporal/021_tbox` (80%), `temporal/009_spanset_ops` (86%), `geo/070_tpoint_spatialrels` (92%), `geo/056_tpoint_spatialfuncs` (79%), `cbuffer/162_tcbuffer_spatialrels` (85%), `pose/105_tpose_spatialfuncs` (88%), `npoint/087_tnpoint_spatialfuncs` (92%), `rgeo/125_trgeo_spatialfuncs` (88%).

Suggested triage order

1. Spatial operator-named-function batch — PR feat(bindings): tgeompoint distance + temporal-position predicates batch #64 already wired the temporal-position predicates (`before`/`after`/`overbefore`/`overafter` on (tstzspan, temporal)). Same pattern needs to extend to `temporal_left/right/below/above` on tnumber × {numspan, tbox} and to all `*_posops` sections per spatiotemporal type.
2. Analytics: DouglasPeucker / simplify family — small surface (4-5 names), self-contained.
3. Aggregates — large but scoped to PR test(parity): 015_span_aggfuncs.test — aggregate surface manifest #21's track. Expanding aggregate infra is a multi-PR effort.
4. Composed transform overloads — `rotate(tgeompoint, double, geometry)` and `scale(tgeompoint, double, double[, double])` are stragglers from PR feat(bindings): tgeompoint affine and derived spatial transforms #65; small follow-up.
5. tgeometry parity for whatever lands on tgeompoint — same overloads, different type tag.

This document is the single source of truth for "what's left." Regenerate after every batch lands so the percentage tracks.

Test plan

• `python3 scripts/parity-audit.py --mdb ../MobilityDB --mduck . --out docs/parity-status.md` succeeds and writes the file.
• Output is reproducible (same inputs → identical file).