Avoid unconditional retries in replicator's http client

In some cases the higher level code from `couch_replicator_api_wrap` needs to
handle retries explicitly and cannot cope with retries happening in the lower
level http client. In such cases it sets `retries = 0`.

For example:

https://github.com/apache/couchdb/blob/master/src/couch_replicator/src/couch_replicator_api_wrap.erl#L271-L275

The http client then should avoid unconditional retries and instead consult
`retries` value. If `retries = 0`, it shouldn't retry and instead bubble the
exception up to the caller.

This bug was discovered when attachments were replicated to a target cluster
and the target cluster's resources were constrainted. Since attachment `PUT`
requests were made from the context of an open_revs `GET` request, `PUT`
request timed out, and they would retry. However, because the retry didn't
bubble up to the `open_revs` code, the second `PUT` request would die with a
`noproc` error, since the old parser had exited by then. See issue #745 for
more.

src/couch_replicator/src/couch_replicator_httpc.erl

@@ -139,12 +139,12 @@ process_response({error, sel_conn_closed}, Worker, HttpDb, Params, _Cb) ->
 %% next request.
 process_response({error, connection_closing}, Worker, HttpDb, Params, _Cb) ->
     stop_and_release_worker(HttpDb#httpdb.httpc_pool, Worker),
-    throw({retry, HttpDb, Params});
+    maybe_retry({error, connection_closing}, Worker, HttpDb, Params);
 
-process_response({error, req_timedout}, _Worker, HttpDb, Params, _Cb) ->
+process_response({error, req_timedout}, Worker, HttpDb, Params, _Cb) ->
     % ibrowse worker terminated because remote peer closed the socket
     % -> not an error
-    throw({retry, HttpDb, Params});
+    maybe_retry({error, req_timedout}, Worker, HttpDb, Params);
 
 process_response({ibrowse_req_id, ReqId}, Worker, HttpDb, Params, Callback) ->
     process_stream_response(ReqId, Worker, HttpDb, Params, Callback);