DOCX inference: References Standards subtree — intentional category breakouts + editor-typo lead-in defeat nesting

[thewrz]

Context

Follow-up to #431 (text-based lead-in nesting). After #431, 1.2 REFERENCES correctly renders its three lead-ins as peers:

A. Abbreviations and Acronyms:   { 1..5 }
B. Definitions:                  { 1, 2 }
C. References Standards:         { ...this subtree... }   ← placement fixed by #431

The subtree under C. References Standards is still parsed incorrectly. Word's own rendering shows the intended structure:

C. References Standards:
   1. Cable:                       ← a numbered category
      a. ICEA S-93-639 / NEMA WC 74-2006 – …
      b. ICEA S-94-649-2021 – …
   2. Cable Sizing:                ← next category
      a. …
   3. Cable Splicing:  /  Cable Test:  /  Cable Terminations:  /  Cable Installation:  …

The parser instead produces the category lead-ins as continuation nodes and the standards as pr4.

Two signals that are hard to correct reliably (why this is deferred)

1. Editor error: the author accidentally typed a stray 1 in front of Cable:, so the first category reads 1 Cable:. That leading 1 (no dot) defeats the Signal-4 marker match (^\d+\.\s), so 1 Cable: classifies as a continuation instead of a numbered category. Correcting this means guessing that 1 is a typo, not content — not reliably inferrable.
2. Designer/editor preference: the author intentionally broke the standards into hand-authored categories (Cable, Cable Sizing, Cable Splicing, Cable Test, Cable Terminations, Cable Installation, …). These categories are only distinguishable as ":"-terminated lead-ins with lettered standards beneath them — recovering that grouping reliably (vs. treating them as flat content) is genuinely ambiguous.

Scope

• Reliably nest the References Standards category subtree (1. Cable: { a., b., … }, 2. Cable Sizing: { … }, …).
• Must be gated on pnpm fixture:snapshot/fixture:diff (0 structural regressions across the corpus) and carry // KNOWN AMBIGUITY tests per the repo rule.
• Repro: docs/references/MANUFACTURER_EXAMPLES/more-parsing-examples-fails.docx, article 1.2 REFERENCES, under C. References Standards.

Until then #431 pins this subtree as a KNOWN AMBIGUITY (its own placement as C. is correct; only its internals are deferred).