v2.6.1

Fixed

• Malformed classic cross-reference subsections are repaired during load. Some PDF producers
  declare a subsection as 1 N while emitting entries for objects 0..N-1, beginning with the
  object-0 free-list head. PdfLibrary previously shifted every offset by one object number and could
  resolve the trailer's /Root to /Pages or /Page instead of /Catalog. The reader now corrects
  the subsection only when both the object-0 free-head signature and the contradictory trailer
  /Size confirm the defect.
• Text extraction now infers spaces from large TJ positioning gaps. PDFs that position each
  word separately without storing literal space characters no longer produce concatenated clauses;
  adjustments above 0.2 em insert a separator while ordinary kerning remains within a word.

Added

• PdfPage.ExtractTextWithQuality() detects implausible Latin text layers. Its
  TextExtractionResult returns the extracted text, printable-ASCII ratio, unexpected-control and
  replacement-character counts, plus IsLikelyReadable. This lets callers route pages whose fonts
  lack a usable /ToUnicode mapping to OCR instead of trusting non-empty but garbled text.