Optimize column listing to not require fetching table properties

[hashhar]

Description

Before this change table metadata was fetched when listing columns which
ended up fetching table properties for JDBC connectors which incurs I/O
in connectors which support them.

Release notes

( ) This is not user-visible or is docs only, and no release notes are required.
( ) Release notes are required. Please propose a release note for me.
(x) Release notes are required, with the following suggested text:

# ClickHouse, SQL Server
* Improve performance of listing columns. ({issue}`23429`)

[hashhar]

while working on this I noticed that we do confusing things in some places.

e.g. ConnectorMetadata#getTableMetadata#getColumns excludes columns like $path in Hive connector but ConnectorMetadata#getColumnHandles includes them. I guess this is to prevent "materializing" those hidden columns in other SQL operations but this makes for confusion. And this is also what prevents applying this optimisation to non-JDBC connectors (or even JDBC connectors too in future if they start returning different columns from getTableMetadata#getColumns VS getColumnHandles.

[hashhar]

For Ignite and Phoenix specifically I'll send a follow-up PR to "unhide" the autogenerated columns we add in case the user doesn't specify primary key (which is required for those systems).

[hashhar]

@kokosing PTAL at the fixups + Add tests to count I/O operations in JDBC connectors (which is the only new commit, it replaces Disable connection reuse in connection counting tests).

[kokosing]

Well done!

[hashhar]

Squashed the fixups, will merge after CI.