pgsql: Fix assorted partition pruning bugs

Fix assorted partition pruning bugs

match_clause_to_partition_key failed to consider COERCION_PATH_ARRAYCOERCE
cases in scalar-op-array expressions, so it was possible to crash the
server easily.  To handle this case properly (ie. prune partitions) we
would need to run a bit of executor code during planning.  Maybe it can
be improved, but for now let's just not crash.  Add a test case that
used to trigger the crash.
Author: Michaël Paquier

match_clause_to_partition_key failed to indicate that operators that
don't have a commutator in a btree opclass are unsupported.  It is
possible for this to cause a crash later if such an operator is used in
a scalar-op-array expression.  Add a test case that used to the crash.
Author: Amit Langote

One caller of gen_partprune_steps_internal in
match_clause_to_partition_key was too optimistic about the former never
returning an empty step list.  Rid it of its innocence.  (Having fixed
the bug above, I no longer know how to exploit this, so no test case for
it, but it remained a bug.)  Revise code flow a little bit, for
succintness.
Author: Álvaro Herrera

Reported-by: Marina Polyakova
Reviewed-by: Michaël Paquier
Reviewed-by: Amit Langote
Reviewed-by: Álvaro Herrera
Discussion: https://postgr.es/m/ff8f9bfa485ff961d6bb43e54120485b@postgrespro.ru

Branch
------
master

Details
-------
https://git.postgresql.org/pg/commitdiff/d758d9702e2f64f08565e18eb6cb7991efa2dc16

Modified Files
--------------
src/backend/partitioning/partprune.c          | 58 +++++++++++------------
src/test/regress/expected/partition_prune.out | 66 +++++++++++++++++++++++++++
src/test/regress/sql/partition_prune.sql      | 14 ++++++
3 files changed, 106 insertions(+), 32 deletions(-)