GreaterWrong: Fix stuck scraper iteration (#198)

* GreaterWrong: Fix setting cooldown on AF/EAF/LW scraper.

* GreaterWrong: Fix iteration to not get stuck on the last item.

Unclear what circumstances cause this to occur,
but sometimes AF returns the item you moved the after date to,
which causes iteration to loop without the date advancing until the job is killed.

This adds a check for the exact case which simply returns,
as well as a more general check for failure to progress which throws.

align_data/sources/greaterwrong/greaterwrong.py

@@ -67,7 +67,7 @@ class GreaterWrong(AlignmentDataset):
     """Whether alignment forum posts should be returned"""
 
     limit = 50
-    COOLDOWN_TIME: float = 0.5
+    COOLDOWN = 0.5
     done_key = "url"
     lazy_eval = True
     source_type = 'GreaterWrong'
@@ -182,16 +182,26 @@ def last_date_published(self) -> str:
     def items_list(self):
         next_date = self.last_date_published
         logger.info("Starting from %s", next_date)
+        last_item = None
         while next_date:
             posts = self.fetch_posts(self.make_query(next_date))
             if not posts["results"]:
                 return
 
+            # If the only item we find was the one we advanced our iterator to, we're done
+            if len(posts["results"]) == 1 and last_item and posts["results"][0]["pageUrl"] == last_item["pageUrl"]:
+                return
+
             for post in posts["results"]:
                 if post["htmlBody"] and self.tags_ok(post):
                     yield post
 
-            next_date = posts["results"][-1]["postedAt"]
+            last_item = posts["results"][-1]
+            new_next_date = posts["results"][-1]["postedAt"]
+            if next_date == new_next_date:
+                raise ValueError(f'could not advance through dataset, next date did not advance after {next_date}')
+
+            next_date = new_next_date
             time.sleep(self.COOLDOWN)
 
     def extract_authors(self, item):