Make NewRouter tolerant to the cluster's degraded state

[nurzhan-saktaganov]

NewRouter returns an error if some (all) cluster replicas are down (or in case of misconfiguration too).

It's ok if some replicas are temporarily unavailable (due to an incident or something else).
However, it is not ok, if we can't start some application that uses go-vshard-router (or create struct Router instance) due to a degraded cluster.

We should be able to work as best as we can even in the worst possible external factors, for example: the availability of cluster replicas is an external factor, that can change over time (a replica can be under maintenance, a network between hosts might be cut off).

But it's ok to return an error due to internal factors, that will not change over time, for example, an invalid config won't fix automatically: an application should be restarted with the corrected config.

[nurzhan-saktaganov]

JFYI, similar issue in go-tarantool tarantool/go-tarantool#372