Remove redundant joins from Views queries

The patch now handles adjust_join() correctly (I think :-)). I made a simple test case: Content type test1 has three single value fields:

- field_field1: text
- field_field2: integer
- field_noderef: noderef to 'story'

Without the optimization, we get the query:

SELECT node.nid AS nid,
   node_data_field_field2.field_field2_value AS node_data_field_field2_field_field2_value,
   node_data_field_field2.nid AS node_data_field_field2_nid,
   node_data_field_field1.field_field1_value AS node_data_field_field1_field_field1_value,
   node_data_field_field1.nid AS node_data_field_field1_nid,
   field_noderef_nid.title AS field_noderef_nid_title,
   field_noderef_nid.nid AS field_noderef_nid_nid
 FROM {node} node 
 LEFT JOIN {content_type_type1} node_data_field_noderef ON node.vid = node_data_field_noderef.vid
 INNER JOIN {node} field_noderef_nid ON node_data_field_noderef.field_noderef_nid = field_noderef_nid.nid
 LEFT JOIN {content_type_type1} node_data_field_field2 ON node.vid = node_data_field_field2.vid
 LEFT JOIN {content_type_type1} node_data_field_field1 ON node.vid = node_data_field_field1.vid

With the optimization, we get the query:

SELECT node.nid AS nid,
   node_data_field_noderef.field_field2_value AS node_data_field_noderef_field_field2_value,
   node_data_field_noderef.nid AS node_data_field_noderef_nid,
   node_data_field_noderef.field_field1_value AS node_data_field_noderef_field_field1_value,
   field_noderef_nid.title AS field_noderef_nid_title,
   field_noderef_nid.nid AS field_noderef_nid_nid
 FROM {node} node 
 LEFT JOIN {content_type_type1} node_data_field_noderef ON node.vid = node_data_field_noderef.vid
 INNER JOIN {node} field_noderef_nid ON node_data_field_noderef.field_noderef_nid = field_noderef_nid.nid

I also fixed the style comments.

This patch has 11 lines of code (including four "}") and 28 lines of comments. ;-)

I just realized a problem. The optimization is working, but the selected fields are getting the wrong AS name: they are being based on the existed queued table name, not the optimized-away joined table name. This needs to be fixed.

Discussion with Earl on IRC concludes that the field renaming issue I mentioned above is not a problem; Views does not guarantee that fields names will be consistent and provides other mechanisms for accessing the data reliably. Therefore, CNR.