fixChangeFeedHangWhenUsingStaleContainerRid

[xinlian12]

Issue:

When query changeFeed with an invalid continuation token (container got recreated in this case, which means the continuationToken uses stale containerRid), SDK will return incorrect result or hang.

Root Cause:

• For container re-created with same RU or same feedRanges -> SDK will reuse the token(lsn) to resume the read from the new container, which could cause missing data

• For container re-created with different RU or different feed Ranges, the feedRange in the continuationToken spans multiple partitions of the new container -> SDK will be in a hang status (due to endless retries).

   1.   During `populateFeedRangeFilteringHeaders` - SDK detected that the feedRange spans multiple partitions, throw `PartitionKeyRangeGoneException`
   2.   `ChangeFeedFetcher.FeedRangeContinuationFeedRangeGoneRetryPolicy`
         -  handle `PartitionKeyRangeGoneException`, it tries to find overlapping ranges based on the continuationToken collectionRid, null list being returned
         -  retry with same feedRange from the continuationToken
         -  repeat step1
  

Fixes:

• For container re-created with same RU or with same feedRanges -> always Populate x-ms-cosmos-intended-collection-rid for changeFeed request, eventually 400/1024 will be bubbled up to customer
• For container re-created with different RU or different feed Ranges, the feedRange in the continuationToken spans multiple partitions of the new container -> SDK will internally create 400/1024 exceptions, refresh container cache once, retry. And if still not successful, then bubble 400/1024 to customer

[azure-sdk]

API change check

API changes are not detected in this pull request.