RS_Within turns antimeridian- and pole-wrapping footprints inside out when reprojecting into the geometry's CRS

`RS_Within` reprojects the raster's (densified) footprint into the geometry's CRS and evaluates the predicate in that flat plane. When the footprint wraps a projection singularity, its interior maps to the *edges* of the target plane and the reprojected ring turns inside out — no amount of edge densification fixes the topology — so SedonaDB answers false for rasters that are truly within the geometry:

**Antimeridian**: an EPSG:3413 square centred on the longitude-180 direction (bbox `(-1330942, 730942, -730942, 1330942)`, corner longitudes ±163.8 and ±180) is truly within the two-lobe region `MULTIPOLYGON(((160 70, 180 70, 180 84, 160 84, ...)), ((-180 70, -160 70, ...)))` — every footprint point lies in a lobe. `RS_Within` returns **false**: the densified footprint's longitudes jump across ±180, and the resulting flat polygon is not within either lobe. (Sedona Spark transforms the geometry into the raster's CRS instead and answers true here.)

**Pole**: the ±1e6 EPSG:3413 pole square is truly within `POLYGON((-180 75, 180 75, 180 90, -180 90, -180 75))` (the latitude-75 cap contains it — footprint latitudes stay above 77). `RS_Within` returns **false**; the pole in the footprint's interior maps to the degenerate latitude-90 edge. (Sedona Spark throws on this one — its corner-only reprojection is apache/sedona#3323 — so on this shape *both* engines fail, in different ways.)

Surfaced by `integration/spark-parity`, where these are xfail-cataloged with truth anchors. A footprint that crosses ±180 or contains the pole in the geometry's CRS needs either evaluation in the raster's CRS (transform the geometry instead, as the intersects path effectively achieves) or split/clamped handling of the seam.

Start by locating the RS_Within implementation and reviewing the integration/spark-parity xfail cases with their truth anchors. Compare its footprint reprojection with the intersects path, then determine how antimeridian and pole-wrapping footprints should be handled. Done means the antimeridian and pole examples return the expected containment results without breaking ordinary cases.

Written by the indexing model from the issue text.