perf(las): skip unneeded TypeScript LAZ layers

[ibgreen]

Goals

• Avoid decoding or allocating LAZ fields that the requested Arrow/COPC output does not use.
• Bring complete-buffer TypeScript PDRF 6-10 parsing onto the optimized direct-Arrow path.
• Document exact LAS/LASzip/PDRF/streaming support without overclaiming by LAS header version alone.

Changes

• Make intensity, classification, and RGB independent selections for direct Point14 output and lock the selection per chunk cursor.
• Skip omitted arithmetic layers and their model allocation for PDRF 6-10.
• Decode complete modern LAZ files directly into Arrow columns for fixed and variable chunks.
• Remove unused COPC intensity/classification arrays and decompression.
• Add selective PDRF 7 parity tests and a COPC render-target benchmark.
• Replace broad LAZ version claims with codec, PDRF, fixture, output, streaming, and COPC support tables; correct LASWriter documentation.

Performance

On ellipsoid-1.4.laz (PDRF 7):

• Complete-buffer TypeScript parse improved from 1.23M to 1.61M output points/s in consecutive benchmark runs.
• Direct positions/RGB output measured 2.19M output points/s versus 1.91M for all exposed Point14 columns.

Benchmark variance was elevated in the second full-suite run. Independently of timing noise, the direct path removes the full raw-record allocation and second point traversal.

Validation

• yarn build
• yarn test-node (403 files passed, 1,918 tests passed)
• yarn test-headless (406 files passed, 1,882 tests passed)
• yarn lint fix
• TS_NODE_LOG_ERROR=true yarn test bench las