Ensure all partitions of a partitioned table has the same distribution.

Ensure all partitions of a partitioned table has the same distribution.

To optimise and simplify XL's distributed query planning, we enforce that all
partitions of a partitioned table use the same distribution strategy. We also
put further restrictions that all columns in the partitions and the partitioned
table has matching positions. This can cause some problems when tables have
dropped columns etc, but we think it's far better to optimise XL's plans than
supporting all corner cases. We can look at removing some of these
restrictions later once the more usual queries run faster.

These restrictions allow us to unconditionally push down Append and MergeAppend
nodes to datanodes when these nodes are processing partitioned tables.

Some regression tests currently fail because of these added restrictions. We
would look at them in due course of time.