This is an old bug which could only surface with the more recent
addition of custom Cypher queries.

The template tries to pre-fetch source data.

It does it for text sources, since they do not support SQL
pushdown, so their data is required for post-processing anyway.

Before this commit, the template also pre-fetched data when none
of the source's targets defined source transformations.

This is overly restrictive and actually wrong.

Custom query targets cannot define source transformations.

If they share a source with a node/rel targets that define custom
source transformations, then the template would crash with a NPE.

This is now fixed. Source data is pre-fetched as long as there is
at least one of its target that does not define any transformation.

The commit also adds another small optimization: if the source
does not match any active targets, the source processing is
skipped completely. Before that, the data could be pre-fetched,
incurring unnecessary data movement.