[arxhr007]

Fixes #22410.

The Adobe Glyph List Specification says that everything from the first period onwards in a glyph name is a suffix (for example c.sc, T_h.alt or one.os), which doesn't change the character(s) of the glyph: https://github.com/adobe-type-tools/agl-specification#2-the-mapping-algorithm

PartialEvaluator._simpleFontToUnicode didn't do that, so the text of simple fonts with such /Differences names was extracted as the character code itself. The glyphs were rendered correctly, but text selection, search and getTextContent returned the wrong characters.

This patch looks up the name without the suffix, if the full name isn't in the glyph list, and then continues with the existing heuristics using that name. This means that c.sc is c, uni00E9.sc goes through the existing uniXXXX handling, and T_h.alt reaches the existing T_h ligature case. A name that starts with a period (.notdef) is left alone.

I did not add the AGL rule for ligature names in general (a_b is ab) since it isn't needed for this issue, and I'd rather keep this change small.

Test:

• New issue22410.pdf (the file from the issue, 3 KB) and an api_spec.js test that expects nicht The café; it fails without the change (ni!ht "e caf#).
• The issue 10640 test expected the same bug. That PDF uses one.os, two.os and zero.os, and its test expected the control characters \x81, \x82 and \x80 where the page has 1, 2 and 0 (for example Directory [\x81] and version \x82.\x80). With this patch the text is Directory [1] and version 2.0, so I updated that expectation.
• The real-world file from the issue (arbeitgeberbescheinigung-max-mustermann.pdf) now has Nein, trifft nicht zu on pages 2 and 3, where it was Nein, trifft ni!ht zu.
• The full unit tests give the same results as master apart from this (the failures that remain are in tests whose PDFs I hadn't downloaded). eslint is clean.

This change was prepared with the help of an AI coding assistant (Devin CLI) and reviewed by me.

[Snuffleupagus]

Sorry, but that issue really does not seem like a good fit for a first-time contributor who appears to just submit fully AI generated code.