Refactor post extraction of VK again to work around their weird behaviours

VK doesn't always return posts in chronological order, so that can't be used to filter out duplicates. Instead, remember the last 1k post IDs and filter using that. This should catch the vast majority of duplicates. (Also, duplicates can't only happen in the geoblocking workaround; sometimes, VK also simply returns the same post again for no obvious reason.)

snscrape/modules/vkontakte.py

@@ -1,4 +1,5 @@
 import bs4
+import collections
 import datetime
 import itertools
 import logging
@@ -100,25 +101,20 @@ class VKontakteUserScraper(snscrape.base.Scraper):
 		# If there is a pinned post, we need its ID for the pagination requests; we also need to keep the post around so it can be inserted into the stream at the right point
 		if 'post_fixed' in newestPost.attrs['class']:
 			fixedPostID = int(newestPost.attrs['id'].split('_')[1])
-			fixedPost = self._post_div_to_item(newestPost)
 		else:
 			fixedPostID = ''
-			fixedPost = None
 
-		lastPostID = float('infinity')
+		last1000PostIDs = collections.deque(maxlen = 1000)
 
 		def _process_soup(soup):
-			nonlocal fixedPost, lastPostID
+			nonlocal last1000PostIDs
 			for item in self._soup_to_items(soup):
 				postID = int(item.url.rsplit('_', 1)[1])
-				if postID < lastPostID:
-					if fixedPost is not None and fixedPostID > postID:
-						yield fixedPost
-						fixedPost = None
+				if postID not in last1000PostIDs:
 					yield item
-					lastPostID = postID
+					last1000PostIDs.append(postID)
 
-		yield from _process_soup(soup.find(id = 'page_wall_posts'))
+		yield from _process_soup(soup)
 
 		lastWorkingOffset = 0
 		for offset in itertools.count(start = 10, step = 10):