write_validated_trait's "byte-identical diff" claim is false for hand-edited trait YAML

[realmarcin]

Found by canarying a round-trip before the #302 edge migration.

src/traitmech/validation/write_validated.py:108 documents its emission options as:

Match the repo's existing yaml emission convention so re-running the helper over
an existing file produces a byte-identical diff.

That does not hold. Loading and immediately re-writing an unmodified file
changes it:

$ python -c "load dissimilatory_iron_reduction.yaml; write_validated_trait(doc, p)"
byte-identical: False        # 10329 -> 10335 bytes, 11 insertions / 9 deletions

Two causes, both cosmetic but diff-noisy:

1. Quote style. note: "Textbook Fe(III)-reducer; ..." (hand-written, double
   quoted) is re-emitted unquoted and wrapped.
2. Line width. safe_dump wraps at its default width (80), which does not
   match where the corpus's hand-edited long strings currently break. Evidence
   snippets and curation_history.changes blocks re-flow.

Why it matters

Any bulk script that touches N trait files via this helper produces a diff where
the real change is buried in reflow churn across every long string in those
files. That is the difference between a reviewable migration and an unreviewable
one — precisely the risk #301 flags for the 366-edge sweep.

The #302 migration therefore uses surgical line-based edits instead, touching
only the lines that actually change.

Fix direction

Either make the claim true or drop it:

• Pass an explicit width= (and possibly default_style) that reproduces the
  corpus's current formatting, then prove it with a test that round-trips every
  file in data/traits/** and asserts zero diff; or
• normalise the whole corpus once through the helper so the claim becomes true
  going forward, and gate it with that same test; or
• soften the docstring to say only that it matches the schema convention, and
  warn that round-tripping a hand-edited file reformats it.

The middle option is the only one that makes the helper safe for bulk use, but
it is a large one-time reformat of 477 files and should be its own PR.