Store ensemble config nodes in an ordered map

We observe different samplings for iteration 0 in regressions tests for
deploys built internally vs deploys built on GitHub Action and then
fetched from PyPI. This commit stores the parameter nodes in an ordered
structure (map) instead of unordered_map aiming at this making the
different builds sample the parameters in the same order.

lib/enkf/ensemble_config.cpp

@@ -23,6 +23,7 @@
 #include <pthread.h>                /* must have rw locking on the config_nodes ... */
 
 #include <unordered_map>
+#include <map>
 #include <string>
 
 
@@ -81,7 +82,7 @@ struct ensemble_config_struct {
   UTIL_TYPE_ID_DECLARATION;
   pthread_mutex_t                                mutex;
   char                                         * gen_kw_format_string;   /* format string used when creating gen_kw search/replace strings. */
-  std::unordered_map<std::string,enkf_config_node_type*>   config_nodes;           /* a hash of enkf_config_node instances - which again conatin pointers to e.g. field_config objects.  */
+  std::map<std::string,enkf_config_node_type*>   config_nodes;           /* a hash of enkf_config_node instances - which again conatin pointers to e.g. field_config objects.  */
   field_trans_table_type                       * field_trans_table;      /* a table of the transformations which are available to apply on fields. */
   bool                                           have_forward_init;
   summary_key_matcher_type                     * summary_key_matcher;