Big Faceless PDF Library full changelog

More optimizations! When profiling or working with the document structure, memory footprint for large documents is down about 20%.
Digitally signing linearized documents broken in the previous release, fixed.
Several improvements to HtmlDerivation class, including some adjustments to the way options are specified in the API. Figures are rasterized to images, ActualText is used from Elements.
ICC profiles used for PDF/A conversion no longer require B2A tables.
Some subtle tweaks to the tag hierarchy rules allowed in PDF/UA-1, and to the exact meaning of the rules over table header "Scope" after cross-testing and discussion with other validators.
Allow a LayoutBox to be drawn twice within a tagged document.
When a single glyph in a font has multiple Unicode values, we were sometimes writing the wrong value to the ToUnicode map, causing later extraction of the text to give the wrong values. Now fixed.
When using the LayoutBox to place mixed right-to-left and left-to-right text on the page, text is now written to the page in logical order so text extraction now works properly even for untagged content.
When extracting Arabic text from a PDF, convert presentation forms back to nominal forms. Extraction is now correct for most tagged content like PDF/UA, although there may be regressions for non-tagged content where text is rendered in reverse order. We hope to have fixes for this in the near future.
Some adjustment to PageExtractor to improve merging of superscript/subscript text with the main row of text, rather than splitting into two rows.
Specific fix for PDFs generated by PDFKit that incorrectly share the same PDF object node for both the AF array and an EmbeddedFile name tree node; this is invalid, but when we repaired one we broke the other. Now split on repair.
Specific fix for PDFs generated by PDFReactor that place "null" in the IDTree
Arlington model would sometimes repeat a warning twice for the same object if it is reachable by multiple p

Minor fixes to ensure the PDF API runs correctly under Java 24.
XMP fixes relating to typed values and qualifiers, added JSON-LD serialization
Locally buffer content being hashed for signature creation, to fix disastrous performance for Amazon CloudHSM v5 signatures.
Cancel unclosed path operations on a loaded-in page before appending to it.
The "fix-invalid-xml" config setting on the Structure Tree now repairs invalid attribute and text values, as well as invalid tag and attribute names.
The serif/sans-serif flag was not being set correctly for OpenTypeFonts.
PDF Association declarations now only refers to WCAG 2.2, so updated OutputProfile feature to match this.
When writing ASN.1 AlgorithmIdentifiers with no parameters, ensure NULL is written only when required.
Viewer: add ability to set minimum length for search term in SearchField

Minor adjustment to PKCS7SignatureHandler.SigningEngine.getTimeStampResp interface, to make it easier to implement a custom SigningEngine
The legacy "Border" attribute on annotations is now supported.
Improved the Font Subsitution algorithm, used when rasterizing a PDF with unembedded fonts.
OpenTypeFont constructor was documented as failing when given a collection and an invalid font name, but it was continuing silently. Now works as specified.
Various changes to make merging of tagged documents (such as PDF/UA documents created by BFO Publisher and Report Generator) easier
Viewer: adjustments to allow a custom SignatureProvider.SignatureState which can reflect site-specific signature validation rules.

Fixed rendering of bitmap images with more than 2^31 samples, and of very large JPX images that are downsampled before cropping
Fixed rendering of files containing unclosed paths before other operations
Previous release built Graal support classes with Java 11, which would break classloaders that foolishly verify classes themselves rather then letting the application test them in a try/catch (Websphere, specifically). Rebuilt with Java 8.
Concatenating PDFs that made use of bfo:sort and bfo:uuid (ie those created by BFO Publisher or Report Generator) had their content reordered randomly, now fixed.
Fixed a very particular error with PDFs containing a particular combination of tokens, which could cause load failures.
A data-structure used when loading fonts from PDFs was causing OutOfMemory errors when used with very large heaps and many threads, despite being based on Weak References. Redid this in a different way using hard references, which seems to have fixed the issue.
Viewer: when the viewer has no option of switching between different layouts, force layout to multi-page view so that navigations between pages is possible.