Use MEST multi-entry indexing in the TRTREE index

[estebanzimanyi]

A temporal column was indexed as a single minimum bounding box per row, so a wiggly or high-extent trajectory matched far more query boxes than it actually overlapped. Construct now indexes each temporal value as up to max_boxes tight per-segment bounding boxes via the MEOS rtree_insert_temporal_split (degenerating to the previous single-box behaviour for instants or max_boxes <= 1); temporal-spatial values are SRID-normalised to 0 before splitting so every produced stbox matches the SRID-stripped query box, and Search deduplicates row ids with a seen-set because multi-entry leaves return the same id once per overlapping segment box. The split bound is exposed as WITH (max_boxes = N), default 8, validated on the MEST selectivity benchmark (false-positive rate 47.2 percent single-box to 13.8 percent at 8). The portfile.cmake change here pins MEOS at the MEST commit only as a TEMPORARY BUILD-BASE so this PR is green in isolation; it deliberately conflicts with the wasm pin work in #142. At finalization this PR drops the portfile edit entirely and inherits #142s pin, which requires MobilityDB #1032 (MEST) to be merged to master alongside #1036 (wasm) so a single master commit carries both; the MobilityDuck merge order is #134, #136, #137, #138, #140, #141, #142, then #139, then this PR. Stacked on the TRTREE crash fix (fix/trtree-index-assertion).

[estebanzimanyi]

Reviewer's quickstart — ~5 minutes

What this PR does in one sentence: switches the TRTREE index from single-entry per row to multi-entry indexing (MEST) — every spatial bbox of a temporal trajectory's component segments becomes an index entry, dramatically improving recall for spatial predicates.

Files (5):

• src/index/rtree_index_create_physical.cpp — the new MEST build path.
• src/index/rtree_index_scan.cpp — adjusted scan to dedup multi-entry hits.
• src/include/index/rtree_module.hpp — type declarations for the new index payload.
• test/sql/index/ — assertions on recall (no false negatives) and result-set deduplication.

Why it matters: previously a single trajectory got ONE bbox in the index — that bbox is loose (covers the whole trajectory's extent), so queries with small predicate polygons returned huge false-positive sets that had to be re-filtered. With MEST, each segment's bbox is indexed separately; a small predicate polygon hits only the relevant segments.

Foundational for: #144 (constant-geometry pushdown — depends on the multi-entry structure existing).

Cross-link: Linux arm64 build needs #161.

Why it's safe to merge: the index becomes strictly more precise — never lower recall (every old entry still maps to the trajectory it came from), never wrong results. Build cost goes up (more entries), query cost goes down (less false-positive filtering).