fix: reset column caches on InvalidQueryException

If a projected query fails with InvalidQueryException (e.g. a column was
dropped from a system table between connections), call projection.reset()
before signalling a reconnect. This ensures the next connection starts
with all caches null and re-discovers available columns via SELECT *,
rather than re-sending the now-invalid projected query.

Locations updated:
- selectPeersFuture: system.peers_v2 onFailure (downgrade path) — changed
  resetPeers() to reset() so the stale peersV2Columns is also cleared
- selectPeersFuture: system.peers onFailure — added reset() on
  InvalidQueryException before propagating the failure
- refreshNodeListAndTokenMap: ExecutionException catch — added reset() when
  cause is InvalidQueryException
- refreshNodeInfo: ExecutionException catch — same

Author: nikagra

driver-core/src/main/java/com/datastax/driver/core/SystemColumnProjection.java

@@ -189,14 +189,6 @@ void populatePeersV2(ResultSet rs) {
     }
   }
 
-  /**
-   * Resets the {@code system.peers} column cache only. Used when downgrading from {@code
-   * system.peers_v2} to {@code system.peers} so that the first peers query re-discovers columns.
-   */
-  void resetPeers() {
-    peersColumns = null;
-  }
-
   /**
    * Resets all column caches so that the next query to each system table sends {@code SELECT *} and
    * re-discovers available columns. Called on reconnection and on schema errors.