Update AI Function Skills based on product updates

[qian-yu-db]

…search

Update ai_parse_document, ai_extract, and ai_classify, and add ai_prep_search (DBR 18.2+). This refresh aligns the skill with the published references and rewrites the custom RAG pipeline around ai_prep_search instead of manual variant_get + explode + md5 chunking.

• ai_parse_document: DBR 17.1+ -> 17.3+; add TIFF/DOC/PPT formats, pageRange option, per-element id/confidence, expanded element type list, 500-page and 100MB limits.
• ai_extract: add object/array schema types, v2.1 options (enableCitations, enableConfidenceScores, version), 150-char field-name and 128k context limits, plus a v2.1 citations/confidence example.
• ai_classify: add version option and 128k context-limit note.
• ai_prep_search: new reference section in 1-task-functions.md (syntax, return shape, SQL + PySpark end-to-end), wired into SKILL.md function table, selection rules, prereqs, and Common Issues.
• 4-document-processing-pipeline.md: rewrite "Custom RAG Pipeline" as parse -> prep -> index -> query; streaming Stage 2 now calls ai_prep_search; Vector Search step explicitly names chunk_to_embed (embedding source) and chunk_id (PK); retain legacy manual-chunking fallback for DBR < 18.2.