Validate tsdb's routing_path

[nik9000]

routing_path's job is to put the same time series on the same node. It
does that by extracting a String directly from the xcontent, hashing it,
and feeding the hash into the shard selection algorithm. I'm happy with
it! But it won't work properly if routing_path matches non-dimension
fields or if it matches non-keyword dimensions. This prevents us from
mapping any fields that aren't keyword dimensions that "line up" with
routing_path.

Let's talk about why routing_path can't do it's job if it matches any
non-dimension fields. Well, imagine routing_path is [dim, foo] and
only dim is a dimension. It'll still hash foo's values into the
routing key. So, say you get documents like:

{"dim": "a", "foo": "1"}
{"dim": "a", "foo": "1"}
{"dim": "a", "foo": "2"}

The third document could be routed to a different shard than the first
two! Which would be a disaster because it'd cut the time series
identified by "dim":"a" into pieces!

Now let's talk about when routing_path matches a non-keyword
dimension. Imagine routing_path is [kwd, int] and we send these
documents:

{"kwd": "a", "int": "1"}
{"kwd": "a", "int": "01"}

Both of these documents belong to the time series identified by
"dim":"a","int":1 but the routing_path code just reads strings so
it'll route them into separate shards. Also bad! Also forbidden by this
change.

[elasticmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)

[imotov]

LGTM in general, added one suggestion.