Support `missing_policy="mask"` on inverse transform

[PGijsbers]

Transferring TODOs to the issue tracker (#172).

automlbenchmark/amlb/datautils.py

Lines 361 to 374 in 98bf554

	def inverse_transform(
	self, vec: str | np.ndarray, **params: Any
	) -> str | np.ndarray:
	"""
	:param vec: must a single value or line vector (array)
	:param params:
	:return:
	"""
	if not self.delegate:
	return vec
	# TODO: handle mask
	vec = np.asarray(vec).astype(self.encoded_type, copy=False)
	return self.delegate.inverse_transform(vec, **params)

In the encoders mask mode, while encoding missing values are temporarily masked as non-missing ones, and then replaced with missing values. During the reverse decoding, as I understand it, we would need to substitute back in those markers, then decode, and then set them to np.nan or None again. (however, that also seems fairly straight forward so I am not 100% why this is then a TODO).