policies: fix dc aware and rack aware policies initialization

[dkropachev]

In cases when nodes are listed not in a proper groups of dc (for DCAwareRoundRobinPolicy) and dc+rack (for RackAwareRoundRobinPolicy).
Policy register only nodes from the last group or dc+rack. So, policy knows only last group of nodes, rest of the nodes considered to be non-existent, until node is restarted, which can be days.

Fixes: #576

Pre-review checklist

• I have split my patch into logically separate commits.
• All commit messages clearly explain what they change and why.
• I added relevant tests for new features and bug fixes.
• All commits compile, pass static checks and pass test.
• PR description sums up the changes and reasons why they should be introduced.
• I have provided docstrings for the public items that I want to introduce.
• I have adjusted the documentation in ./docs/source/.
• I added appropriate Fixes: annotations to PR description.

[Lorak-mmk]

Personally I think such comprehensions / unpacking (or w/e is this called) is much less readable than something simpler like

for dc, dc_hosts in groupby(hosts, lambda h: self._dc(h)):
    self._dc_live_hosts.setdefault(dc, set())
    self._dc_live_hosts[dc].update(dc_hosts)

I'd prefer it to be changed, but I won't hold this PR over that.

[Lorak-mmk]

Btw if we are staying with the current version, let's change .get(dc, []) to .get(dc,set()).
It won't change how the code works but makes it imo easier to reason about - _dc_live_hosts is a dict str -> set of hosts, so let's keep it that way.