Remove the deprecated load_*/save_* aliases

[skearnes]

The load_*/save_* rename (#818, #833) left forwarding shims that warn and delegate. They shipped in v0.6.30 and have now been available through two minor releases, so they go:

removed	replacement
ord_schema.parquet_dataset (whole module)	ord_schema.parquet
parquet.write_dataset	parquet.save_dataset
parquet.read_dataset / read_metadata / read_footer / read_reaction	parquet.load_dataset / load_metadata / load_footer / load_reaction
message_helpers.write_message	message_helpers.save_message
message_helpers.write_dataset	ord_schema.datasets.save_dataset
templating.read_spreadsheet	templating.load_spreadsheet
resolvers.name_resolve	resolvers.resolve_name

The deprecated --input alias for --input_pattern goes too, from both validate_dataset.py and build_dataset.py. ord-data's validation.yml still passes --input, but it runs these scripts out of a pinned ORD_SCHEMA_TAG checkout, so it keeps working until that pin moves and switches to --input_pattern at the same time.

Nothing inside ord-schema used any of them, and no test asserted the warnings. The two guides that named write_message and name_resolve now name save_message and resolve_name.

Why the module shim was the urgent one

ord_schema.parquet_dataset forwards through a module-level __getattr__ returning Any. That means a type checker resolves the module fine — it exists — and then accepts every attribute reached through it, including ones that do not exist:

parquet_dataset                     → <module 'ord_schema.parquet_dataset'>
parquet_dataset.DatasetWriter       → Any
parquet_dataset.NoSuchThingAtAll    → Any

So a caller on the alias got no type checking of the API surface at all, and would have kept type-checking clean right up until it failed at runtime. That is exactly what happened in ord-data, where convert_to_parquet.py had been on the alias since the rename with ty reporting nothing.

Callers elsewhere

• ord-data — convert_to_parquet.py and the README conversion example, fixed in ord-data#263.
• ord-interface — ord_interface/editor/py/serve.py uses templating.read_spreadsheet and resolvers.name_resolve. Migrated in ord-interface#213, merged.
• ord-app — unaffected. Its write_message and name_resolve_cached are its own functions, and it pins <0.7.

This is a breaking change for anyone still on the old names, so it wants the next minor version rather than a patch.

Verified locally: ruff check, ruff format --check, ty check, all pre-commit hooks; 534 tests pass.

🤖 Generated with Claude Code

Greptile Summary

Removes deprecated API and CLI compatibility aliases.

• Deletes the ord_schema.parquet_dataset forwarding module and deprecated load/write/read function aliases.
• Removes the deprecated --input spelling from the dataset build and validation CLIs.
• Updates guides to use save_message and resolve_name.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
ord_schema/message_helpers.py	Removes deprecated message and dataset write aliases after repository callers migrated to canonical save APIs.
ord_schema/parquet.py	Removes deprecated read/write forwarding functions without changing the canonical Parquet implementation.
ord_schema/parquet_dataset.py	Deletes the deprecated module-level forwarding shim for ord_schema.parquet.
ord_schema/resolvers.py	Removes the deprecated name_resolve alias while retaining resolve_name unchanged.
ord_schema/templating.py	Removes the deprecated spreadsheet-loading alias while retaining load_spreadsheet.
ord_schema/scripts/build_dataset.py	Removes the deprecated --input CLI alias and keeps --input_pattern as the required canonical argument.
ord_schema/scripts/validate_dataset.py	Removes the deprecated --input CLI alias and keeps existing validation behavior unchanged.
docs/guides/compound_identifiers.rst	Updates resolver references and links to the canonical resolve_name API.
docs/guides/templates.rst	Updates template-writing guidance to reference save_message.

_{Reviews (3): Last reviewed commit: "Merge remote-tracking branch 'origin/mai..." | Re-trigger Greptile}

Context used (3)

• Knowledge Base — Data I/O: proto serialization, Parquet, and units
• Knowledge Base — Dataset Scripts
• Knowledge Base — Validation, Resolution, and Update Pipeline