Extension security improvement: Add support for extensions with an owned schema

Writing the sql migration scripts that are run by CREATE EXTENSION and
ALTER EXTENSION UPDATE are security minefields for extension authors.
One big reason for this is that search_path is set to the schema of the
extension while running these scripts, and thus if a user with lower
privileges can create functions or operators in that schema they can do
all kinds of search_path confusion attacks if not every function and
operator that is used in the script is schema qualified. While doing
such schema qualification is possible, it relies on the author to never
make a mistake in any of the sql files. And sadly humans have a tendency
to make mistakes.

This patch adds a new "owned_schema" option to the extension control
file that can be set to true to indicate that this extension wants to
own the schema in which it is installed. What that means is that the
schema should not exist before creating the extension, and will be
created during extension creation. This thus gives the extension author
an easy way to use a safe search_path, while still allowing all objects
to be grouped together in a schema. The implementation also has the
pleasant side effect that the schema will be automatically dropped when
the extension is dropped.

One way in which certain extensions currently hack around the
non-existence of this feature is by using the approach that pg_cron
uses: Setting the schema to pg_catalog and running "CREATE SCHEMA
pg_cron" from within the extension script. While this works, it's
obviously a hack, and a big downside of it is that it doesn't allow
users to choose the schema name used by the extension.

PS. I have never added fields to pg_catalag tables before, so there's
a clear TODO in the pg_upgrade code related to that. If anyone has
some pointers for me to look at to address that one that would be
helpful, if not I'll probably figure it out myself. All other code is
in pretty finished state, although I'm considering if
AlterExtensionNamespace should maybe be split a bit somehow, because
owned_schema skips most of the code in that function.