Prevent assert failure in scheduling logic

This fixes a bug introduced in commit f1cf6d2d
Without this change, a job with a partition related reason for not running
(e.g. MaxNodes) plus some other reason (e.g. dependency), could after satisfying
the second reason (e.g. dependency), have the select_nodes() function executed
and return SUCCESS on an allocation request, but it would not actually be
allocation resources since select_nodes() would interpret the request as a
test to see if it could ever run (e.g. will_run).
Bug 263