- Correct buffer re-allocation

- buildbot test failure fixes: don't try to get more records from SimpleIndexReader
  if we've already got EOF for it.

- address the review feedback (class/var renames)
- fix wrong test result (bug in buffer refill logic)

- One iterator class
- Switch back from state automaton into two-nested-iterators approach..

- change dsmrr_next_from_index() to a switch-based state automaton-like structure.

- Switch from one bi-directional buffer class to two 
  virtual inheritance-based forward and backward buffer classes.

  for writing and reading

- better comments
- rename variables to better reflect their meaning

- First part of the fix: enable Early NULLs filtering to work when WHERE clause is present

BUG#628785: multi_range_read.cc:430: int DsMrr_impl::dsmrr_init(): Assertion `do_sort_keys || do_rowid_fetch' failed 
- Make Ds_MrrImpl::check_cpk_scan() follow the execution code' logic: don't
  do MRR scans on clustered PK when mrr_sort_keys=off.

- When find_all_keys() checks which table columns are needed for table scan
  that is done before the sorting, it should also analyze pushed index condition.
  This is achieved by remembering/checking pre-index-pushed condition.

- More test coverage

- Relax overly-strict assert added in previous commit
- Run generic MRR testsuite for Maria engine, too.

- Address review feedback, step 1 

- Make testcase stable 

- don't allocate space for rowid buffer when we don't really need it.
- fix buffer iterator