fix: improve geography/geometry parser robustness

[dhensby]

Summary

Fixes three bugs in the geography/geometry binary parser (lib/udt.js) that caused crashes on certain spatial data inputs.

Bugs Fixed

1. H flag (IsLargerThanAHemisphere) collision with P flag — The H flag was checked against bit 3 (1 << 3), which is the same bit as the P (single point) flag. This meant v2 geography data with the H flag set would be incorrectly parsed as a single point. Fixed to use the correct bit 5 (1 << 5).

2. No bounds checking on buffer reads — The parser performed no validation that sufficient data remained in the buffer before reading. Truncated or corrupt spatial data would produce cryptic RangeError: Attempt to access memory outside buffer bounds errors. Added an ensureBytes() helper that throws a clear 'Spatial data is truncated' error before every read section.

3. Segment count check off-by-one — The check for whether segments existed used buffer.position < buffer.length, which would allow 1-3 trailing bytes to trigger a 4-byte read that overflows the buffer. Fixed to buffer.position + 4 <= buffer.length.

Tests Added

16 new unit tests covering:

• Single point (P flag) for geography and geometry
• Single line segment (L flag)
• Z values, M values, and Z+M combined
• v2 with circular arc segments
• v2 with multiple segment types
• v2 H flag (IsLargerThanAHemisphere) — validates the bit 5 fix
• MultiPolygon and GeometryCollection (nested shapes)
• Null geometry (SRID = -1)
• Empty geometry (zero points/figures/shapes)
• Truncated data: too-short header, truncated points, truncated figures, truncated Z data

Fixes #322