Concurrent walk test

[schomatis]

All the numbers and estimations need to be checked later. The purpose of this review is to check if this was the design requested in #99 (comment).

The main issue here is that for TestHAMTEnumerationWhenComputingSize to pass we now need a low-level understanding of the dag.Walk function when called with the Concurrent option (an example of this is that we are copying one of its internal variables, defaultConcurrentFetch, in order to anticipate how many DAGService.Get() requests will roughly be issued).

Another example of the above is that confusingly WalkDepth is not necessarily a DFS. "Depth" here just means that the depth of the link being visited is passed to the callback but in practice the traversal is:

• for the sequentialWalkDepth is a recursive algorithm and hence DFS approach,
• but for parallelWalkDepth a FIFO queue is used and hence we have a BFS approach.