Adds v2 SpanStore and a bridging adapter to v1

[codefromthecrypt]

This represents the minimal read apis needed for Zipkin

As discussed earlier on gitter w/ @basvanbeek, we no longer do
adjustment at storage layer. Essentially, all operations are "raw",
which reduces the responsibility of implementors. This also midly
changes the getTrace call, disallowing it from returning null (as
empty will suffice).

This adds SpanStore which returns Call objects as opposed to replicating
signatures for synchronous and asynchronous invocations. Unlike before,
this does not support "adjustments" rather assumes they happen at a
layer above storage. This decision refines the storage query api to the
following:

Call<List<List<Span>>> getTraces(QueryRequest request);

Call<List<Span>> getTrace(long traceIdHigh, long traceIdLow);

Call<List<String>> getServiceNames();

Call<List<String>> getSpanNames(String serviceName);

Call<List<DependencyLink>> getDependencies(long endTs, long lookback);

A later change might replace the getTrace call with a hex string
parameter.

[codefromthecrypt]

PS once this is in, the v2 http api (openapi/swagger etc) can be implemented. Later, we can formalize and make the java libraries not-internal.

[codefromthecrypt]

ok all apis are here, and I'll raise two other PRs: one to add the http api and another to implement with elasticsearch. There's still a solid amount of work on this PR, yet, which is to backfill tests and make a v2 replacement for SpanStoreTest

One thing you'll notice is I made a copy of the bridging read api from elasticsearch to here. This is to decouple change. The elasticsearch copy will be deleted later.

cc @anuraaga

[codefromthecrypt]

OK backfilled almost all tests. nearly there