roundrobin should handle a case if selected endpoint returns error

[denis-tingaikin]

Actual behavior

Currently, roundrobin fails if the selected endpoint is busy or returns an error https://github.com/networkservicemesh/sdk/blob/master/pkg/networkservice/common/roundrobin/server.go#L63

OR The same problem with other scerio:

1. bad NSE registered with ns-1
2. NSC requested NSE with ns-1
3. working NSE registered with ns-1 (in parallel with step 2)

Actual: NSC fails due to roundrobin selects bad NSE and not tries to select working NSE.

Expected behaviour

roundrobin should try to select next endpoint if the previous one fails.

Potential solution

1. rounrobin should iterate over all Candidates in round-robin sequence.
2. discovery should try to call next with new candidates if they've registered during request AND previous candidates fail AND ctx is not done.