Document transliteration data struct format

[skius]

The zero-copy format of rule based transliterators (introduced in #3775) should be documented more.

• How rule encoding works with private use code points
• How cursor offsets work with special replacers (backrefs, function calls)
• How backrefs are encoded ($n is n code points after the last code point that maps to a VarZeroVec in VarTable, e.g., if the only specials are a UnicodeSet, a segment, and a backref $1, then $1 would be the third private use code point (first is segment, second is UnicodeSet)))
• How anchors are encoded (two reserved, hardcoded private use values for start and end)
• What the {id/rule}_group_list fields are and how they need to be interpreted

[skius]

Include this in documentation: #3891

[skius]

See the design doc: https://docs.google.com/document/d/1KoqwKKD7qu5Y3fW1ONRwByk-IGW_QgsUr6UwyVcyRjQ/edit?resourcekey=0-izQovenrD6PsOa_GNW1SLw