Currently Table.cache() will result in a new copy of the data being stored in the backend, even if the data is already a "concrete" table in the backend.

Ideally if a table is already concrete (backed by a physical table, not a view, in the corresponding backend) then table.cache() would be a no-op. This would better enable writing generic functions that make use of .cache without unnecessarily duplicating data.

t = con.table("my_table")  # a concrete table

_ = t.cache()  # a no-op

_ = t.mutate(foo=t.bar + 1).cache()  # not a no-op, since `t` isn't a physical table

t = con.table("some-view")  # a view

_ = t.cache()  # not a no-op, since `t` is a view