For some reason, the code reading answers was using imap() function to
read them _in order_ they were sent. This could cause answer buildup
in an internal multiprocessing queue, and eventually the program logged
multiple 'Received xxx answers' lines at once, which made the status
reporting useless.

It was like that since d422292d, and
this commit claims it reduced storage requirements by factor 25.
The problem is that the commit did three changes at once:
- it moved txn.put() into central place in the main thread
- it coalesced all writes into one huge transactoin
- it changed imap_unordered() to imap() call

In my tests today, centralized txn.put() in the main thread in
combination with one huge transaction and imap_unordered() increases
storage size by 5 %, and that does not seem worth incorrect status
reporting.

Tested with LMDB version 0.9.29.