fix(chunker): image-only pages silently skipped, leaving sections with zero chunks but visible in TOC #276

Summary

chunker.py's _section_page_text() filters out pages where page.text.strip() is falsy. Pages with only images (no extractable text) produce empty page.text, are silently skipped with no warning, and the resulting section has zero chunks — yet it still appears in the document's TOC and routing responses, returning zero search results.

Root Cause

ingestion/chunker.py around line 95:

def _section_page_text(doc: ParsedDocument, entry: TocEntry) -> Iterator[tuple[int, str]]:
    for page in doc.pages:
        if page.page_num < entry.page_start or page.page_num > entry.page_end:
            continue
        if not page.text.strip():   # ← image-only pages silently skipped
            continue
        yield page.page_num, page.text

No log entry is emitted when a page is skipped. A section spanning 5 pages where pages 2–4 are full-page images produces a chunk from page 1 and page 5 only — without the caller knowing that 60% of the section's content was silently ignored.

Failure Scenario

1. A regulatory document has a section "Compliance Matrix" spanning pages 10–15.
2. Pages 10–14 are full-page tables rendered as images (no text extractable).
3. Page 15 has a one-line summary.
4. chunk_section produces one chunk from page 15.
5. The section appears in TOC as "Compliance Matrix (pages 10-15)".
6. A query "What is in the compliance matrix?" returns only the one-line summary from page 15.
7. 85% of the section's content is permanently invisible, with no indication to the operator.

Proposed Fix

Log a warning when pages are skipped:

if not page.text.strip():
    logger.warning(
        "Page %d in section '%s' has no extractable text (image-only?) — skipping",
        page.page_num,
        entry.title,
        extra={"doc_id": doc.doc_id, "section_id": entry.section_id},
    )
    continue

Expose skipped_pages count in IngestResult so operators can detect image-heavy documents:

@dataclass
class IngestResult:
    ...
    skipped_image_pages: int = 0