refactor: remove duplicate structured field helper - #183

Closes awslabs#213 for everything currently mergeable. awslabs#159 and awslabs#183 have shipped
with no entry; awslabs#146 and awslabs#198 are still open, so their entries belong with
those PRs rather than here.

awslabs#159 gets three entries, two Fixed and one Changed. Verified each against the
merge parent (b78c1f0) rather than trusting the issue text, which turned out
to describe the export shape wrongly:

    to_json_schema()      before {"type": "string"}
                          after  {"type": ["string", "null"]}
    model_json_schema()   before {"type": "string"}
                          after  {"anyOf": [{"type": "string"}, {"type": "null"}]}

The issue said both emit `anyOf`. Only `model_json_schema()` does -- that is
Pydantic rendering `Optional[T]` -- while `to_json_schema()` uses the list
form. The entry documents both, since a reader feeding either into a validator
needs the one they actually call.

Also confirmed the other claims held before writing them up: explicit `null`
now accepted for an optional schema-built field (ValidationError before),
required fields still reject `None`, re-import still idempotent, and
`is_structured_field_type()` now returns True for `Optional[SomeModel]` where
it returned False. That last one needed a FieldInfo rather than a bare
annotation to reproduce -- passing the annotation returns False on both sides,
which would have read as the claim being wrong.

awslabs#183 deliberately gets no entry: 44 lines of a duplicate private helper
removed, no runtime change. The issue asked for a decision rather than a
default, so the convention is now written down in RELEASING.md -- entries are
for changes a user can observe, internal refactors are git log's job. 0.6.0's
notes already followed this; it just had not been stated.

Added a CHANGELOG checkbox to the PR template, phrased so "not needed" is an
explicit choice rather than an omission. Every gap in awslabs#213 is a PR that passed
review without one, which points at the process rather than the authors.

Suite 1746 passed, 2 skipped; mkdocs clean; the template's relative link to
RELEASING.md resolves.