Transfers: implement source selection strategies

Try to be more generic in the source filtering and selection code.
Introduce a SourceRankingStrategy class, which has an `apply` method
to actually apply the strategy to a source and decide what to do with
a source. Some strategies are filter-only: i.e. they only decide if
a source has to be ignored or not. Others are also ranking strategies:
they can return an integer which defines the cost of the source
according to this strategy. The order of ranking strategies is
important: it defines the order in which the costs will be compared to
sort the sources.

This change will definitely introduce some computational and memory
overhead, but I hope it will be low. Instead of hard-coded `filter()`
calls which are done once per request, now we have multiple object
traversals and function calls for each source of each request.

lib/rucio/core/request.py

@@ -56,8 +56,8 @@
 
 
 class RequestSource:
-    def __init__(self, rse_data, ranking=None, distance=None, file_path=None, scheme=None, url=None):
-        self.rse = rse_data
+    def __init__(self, rse: RseData, ranking=None, distance=None, file_path=None, scheme=None, url=None):
+        self.rse = rse
         self.distance = distance if distance is not None else 9999
         self.ranking = ranking if ranking is not None else 0
         self.file_path = file_path
@@ -558,7 +558,7 @@ def list_and_mark_transfer_requests_and_source_replicas(
         if replica_rse_id is not None:
             replica_rse = rse_collection[replica_rse_id]
             replica_rse.name = replica_rse_name
-            source = RequestSource(rse_data=replica_rse, file_path=file_path,
+            source = RequestSource(rse=replica_rse, file_path=file_path,
                                    ranking=source_ranking, distance=distance, url=source_url)
             request.sources.append(source)
             if source_rse_id == replica_rse_id: