context.entities should be keyed based on the entity long name

[effigies]

Noticed while reviewing #71: entities["sub"] is populated, rather than entities["subject"].

The schema consistently uses the long name, not the short name. This means we'll need to look up the mapping from schema.objects.entities, and we should probably do that once per schema.

I recently introduced childmindresearch/bids2table#74 as a method for mapping schema specifiers onto an adapter class that extracts the bits the library actually needs. The basic idea is:

SchemaSpec: ty.TypeAlias = Namespace | os.PathLike[str] | str | None

def load_bids_schema(spec: SchemaSpec = None) -> BIDSSchemaAdapter: ...

def public_function(*args, schema: SchemaSpec = None) -> ReturnType:
    return _private_function(*args, adapter=load_bids_schema(schema))

def _private_function(*args, adapter: BIDSSchemaAdapter) -> ReturnType: ...

BIDSSchemaAdapter is never exposed as a public API, but is able to be passed around cheaply throughout private functions.

We can start by replacing the datatype-to-modality map:

python-validator/src/bids_validator/context.py

Lines 54 to 63 in e85ad33

	_DATATYPE_MAP: dict[str, str] = {}
	def datatype_to_modality(datatype: str, schema: Namespace) -> str:
	"""Generate a global map for datatype to modality."""
	global _DATATYPE_MAP
	if not _DATATYPE_MAP:
	for mod_name, mod_dtypes in schema.rules.modalities.items():
	_DATATYPE_MAP |= dict.fromkeys(mod_dtypes['datatypes'], mod_name)
	return _DATATYPE_MAP[datatype]

And a short-to-long entity name map would be good to have.

Need to consider how to actually thread this through. I suspect we attach the adapter to the Dataset object, so it's always available to the individual context.