#37 NeverEnding reader postpones last message

[ABMC831]

Release notes:

• split never-ending / ending reader implementation thus fixing the bug

Logic of when to fetch next batch, is too different between reader nodes,

for ending, it needs to happen AFTER item BEFORE return, so hasNext can answer

for never-ending, it needs to happen BEFORE item, since it can take forever, which should not be done inside hasNext, which is always true

Also timeout of 3 seconds is just not enough for our poor cluster to not-return empty batch on the first fetch

Also changed default to Ending, so samples can actually be run without blocking on first ocurence of reader

closes #37