feat: implement real relation type [WIP]

[ingomueller-net]

No description provided.

[ingomueller-net]

I am still unsure about what assembly format is best. We have the following possibilities:

%0 = named_table @t1 as ["a"] : <si32>  // no typename by default
%0 = named_table @t1 as ["a"] : !substrait.relation<si32>  // always fully qualified name
%0 = named_table @t1 as ["a"] : relation<si32>  // short pseudo-typename

If no type name is the default, we have to rely on the reader's knowledge that the ops always return relations. If we always print the fully qualified name, some ops get quite verbose (set ops can have several operands plus the result, which would all repeat that it's a relation). The short pseudo-typename would really be a keyword, which (1) would need to be defined for each op and (2) wouldn't color as a type in IDEs. Any opinions, @jpienaar or @dshaaban01?