Refuse or carry the edits a re-read would undo

[ww-mw]

Four findings from a test sweep of the unified table-edit path, all one shape: the
table accepted an edit, showed it, and the next read of the file spelled something
else.

• A rename of a catalogued Architectural Data entry demoted it. System Composer
  keeps its type catalog in a separate part of the dictionary that references entries
  by name (a JSON object in a textual .sldd, a zipped MF0 XML member in a
  compressed-binary one), so a rename left the catalog defining a name nothing carries
  and a struct type re-read as a plain data interface — in the file, on both formats.
  New parser/ScCatalog.ts holds that vocabulary; SectionNode._entryRenamed moves the
  key in memory (load-bearing for a narrow repaint, which rebuilds an entry from a
  record and classifies it through the catalog), and the scan/rename/apply helpers are
  exported so the host can patch the file — it owns the document text and the zip
  members; this package can splice neither.
• A rename of a struct/object/.mat struct array left its elements labelled after the
  old name. The label was baked at parse time; it is now derived from an
  ElementSubscript plus the parent's current displayed name.
• A Simulink.Parameter's Value row offered a rename serialize discarded — the
  property is written under the key Value whatever the node is called. The row is a
  class property bag, so it says so, and setProperty asks nameEditable first.
• A Description typed onto a plain variable or a struct went nowhere — those
  serialize as {name, metadata, value}. descriptionEditable joins the
  nameEditable/valueEditable family and reaches the row, the property inspector and
  setProperty.

Tests: test/scCatalog.test.ts (both storage forms, the rename sites, and the sites a
rename must NOT touch), test/editsTheFormatCannotKeep.test.ts,
test/elementLabelFollowsName.test.ts, plus new fixtures arch_binary.sldd /
arch_binary_as_text.sldd generated by make-fixtures.mjs. Full suite: 4134 passing.