Add PPTX slide and speaker-notes redaction

[maziyarpanahi]

Summary

Adds PPTX ingestion and PHI redaction for slide shapes, table cells, and speaker notes through the multimodal ingest contract.

Changes

• Extract slide, table-cell, grouped-shape, and speaker-notes text with per-slide and run-level character-offset provenance.
• Project detected spans back to PowerPoint runs with PHI-safe offset, source-location, confidence, and digest metadata.
• Optionally write redacted text into a copy of the deck while preserving unaffected run formatting.
• Register the .pptx handler, expose the public API, document usage, and add the permissive-license dependency metadata.
• Add synthetic offline acceptance tests for extraction, offset round-tripping, registry dispatch, and write-back.

Root cause

The multimodal ingest contract supported DOCX but had no PPTX adapter, leaving slide text and speaker notes outside the text redaction path.

Validation

• .venv/bin/python -m pytest tests/unit/multimodal/test_pptx_redact.py tests/unit/multimodal/test_multimodal_extra.py -q — 4 passed, 1 skipped.
• Scoped Ruff format check passed for all changed Python files.
• Scoped Ruff lint check passed for all changed Python files.

Synthetic offline data only; embedded images remain out of scope.

Closes #860

[maziyarpanahi]

Integrated via #2687 after exact-head validation. The batch passed 54 focused tests (1 skipped), Ruff checks and format verification, and a strict documentation build; the resulting master tree was verified exactly.