[python] Support ARRAY BLOB in multimodal payload reads

[TheR1sing3un]

Purpose

read_blobs() and stream_blobs() reject ARRAY<BLOB> columns even though the storage reader supports them. Reading another BLOB column also leaves the array's descriptors in the scalar output.

Recognize ARRAY BLOB columns, fetch their elements in the same coalesced range read as scalar and MAP BLOBs, and restore row-aligned lists. Preserve element order, duplicates, null arrays, empty arrays, null elements, and empty or inline payloads. Projection excludes unrequested ARRAY BLOB columns from scalar output.

Tests

• Multimodal table and contiguous-window suites: 124 passed.
• Focused BLOB reads on PyArrow 16: 7 passed.
• Coverage includes mixed scalar/MAP/ARRAY columns, coalesced reads, filtering, projection, row IDs, limits, and empty results.
• Flake8, changed-file license checks, Python 3.6 grammar checks, and git diff --check passed.

[JingsongLi]

Requirement fit: SUPPORTED (triage: GO)
Implementation: CLEAN

ARRAY<BLOB> is a useful missing case for multimodal reads. The change keeps array element order, null and empty arrays, duplicate elements, and row alignment while fetching bodies through the existing coalesced read. The end-to-end tests also cover projection, filtering, row IDs, streaming, and mixed scalar/MAP/ARRAY columns. I found no blocking issue in the changed path.