replicator memory management and buffer flush calculation updates

* new should_flush fun considers ndocs, nattachments, memory in making decision
* memory utilized by attachment receivers is accounted for
* download attachments using standalone connections instead of conn pool.  This
  prevents a document request from getting stuck behind a huge attachment, which
  would prevent us from triggering a buffer flush in time.  We also consider the
  memory utilization of the standalone ibrowse connection in should_flush


git-svn-id: https://svn.apache.org/repos/asf/couchdb/trunk@775507 13f79535-47bb-0310-9956-ffa450edef68

0 files changed, 0 insertions, 0 deletions