SpanR.setTag() throws NPE when using null key/value

[kcirone]

This is different behavior than, for example, Jaeger, due in part to Jaeger using a HashMap instead of a ConcurrentHashMap, the latter of which does not accept null values.

The fix would be to switch from using ConcurrentHashMap (which seems sensible given that spans really ought to be measuring operations on a single thread) OR at the very least checking for null. (FWIW the OpenTracing spec is mute on the subject of null tag values.)