refactor(extensions): move example YAML files out of the extensions folder

[nielspardon]

Problem

The core extensions/ folder mixes official function extensions (functions_*.yaml) with example/placeholder files:

• type_variations.yaml (extension:io.substrait:type_variations) — example type variations, no functions
• extension_types.yaml (extension:io.substrait:extension_types) — the point/line user-defined types used as the docs' UDT example

Because they live alongside the real extensions, consumers ingest them inconsistently. This surfaced in duckdb-substrait-extension#239 (issue #214), where the now-removed unknown.yaml placeholder was expanded into bogus catch-all function registrations that clobbered valid ones.

Note: the third placeholder, extensions/unknown.yaml, was already removed in #1081, which replaced it with the built-in unbound type. This PR was rebased accordingly and no longer touches it.

Filtering behavior across the ecosystem for the two remaining files is inconsistent:

Consumer	Strategy	type_variations	extension_types
DuckDB (post-#239)	allowlist functions_*.yaml	skip	skip
substrait-python	allowlist functions*.yaml	skip	skip
substrait-java	hardcoded curated list	excluded	loaded
substrait-go	load all	loaded	loaded
substrait-rs (+packaging)	load all; rs skips by URN	loaded (inert)	skip (stale)

Change

Move the two example files to site/examples/extensions/ (the existing, schema-validated, docs-embeddable examples folder) and rename their URNs to the extension:example:* convention already used there. The core extensions/ folder now contains only official functions_*.yaml extensions.

References to extension:io.substrait:extension_types are updated to extension:example:extension_types in the docs (extensions/index.md, types/type_parsing.md), the distance_functions.yaml example dependency, and the parser test fixtures in tests/coverage/test_coverage.py. The type_variations URN is not referenced anywhere else.

Also adds a .DS_Store entry to .gitignore.

No tooling/config changes were needed: the coverage tests and function-doc generator already filter to functions_*, check-examples now validates the moved files against the same schema check-extensions used, yamllint runs repo-wide, and the mkdocs --8<-- include root is unchanged.

Verification

• pixi run yamllint — clean
• pixi run check-jsonschema — examples + extensions + dialects all ok
• pixi run pytest tests/test_extensions.py tests/coverage/test_coverage.py — 128 passed
• pixi run mkdocs build --strict — builds, includes resolve
• git grep extension:io.substrait:{type_variations,extension_types} — no matches remain

Open questions for maintainers

1. Is extension_types an example or an official extension? Unlike type_variations, it's genuinely load-bearing today (substrait-java and substrait-go load its point/line types). This PR treats it as an example; if we'd rather ship it officially, it should stay in / return to extensions/ under an official URN and only type_variations moves.
2. Where should examples live? This PR uses the docs tree (site/examples/extensions/, not shipped in the release artifact). An alternative is shipping examples under extensions/examples/ so they travel with the extensions tree for downstream testing — the file moves and URN renames are identical; only the target directory and the check-extensions glob differ.

Downstream follow-ups (separate PRs)

Consumers that vendor the extensions folder will need updates when they next bump, since both files moved and the extension_types URN changed:

• substrait-java: loads extension_types.yaml + defines the EXTENSION_TYPES URN constant; would lose the point/line UDTs unless it vendors the example from the new location.
• substrait-go: loses extension_types types when it re-embeds.
• substrait-rs: the URN skip for extension_types becomes dead (its cited bug docs: clarify u! prefix requirement for user-defined types in YAML #935 is already fixed).
• substrait-packaging: generate_extensions.sh copies top-level extensions/*.yaml, so the moved files drop out naturally — no script change.
• DuckDB (feat(proto): add write relation to support multiple outputs #239) and substrait-python: functions* allowlist — no impact.

🤖 Generated with AI

---

This change is