[TASK] Add capability to EmbeddingTable to hold multiple features that share id-encoding

[marcromeyn]

As of now a EmbeddingTable holds a single feature which doesn't allow for shared embedding-tables. To support this we propose to add the capability to a table to add features as follows:

item_id_col = ...
user_item_history = ...

table = EmbeddingTable(64, item_id_col, user_item_history)

table.add_feature(col)

table.schema # -> Schema(item_id_col, user_item_history)

# Ways to call this
table(tf.Tensor([1,2,3])) # Tensor[3, 64]
table({"item_id": tf.Tensor([1,2,3])}) # Dict[str, Tensor[3, 64]]
table({
	"item_id": tf.Tensor([1,2,3]),
	"user_item_history": tf.RaggedTensor([1], [2, 4], [3])
}) # Dict[str, Tensor[3, 64]]