Remove placeholder edges from the flow network

[wkearn]

Resolves #166

This change renames flow_routing_targets to
flow_routing_d8_targets and fills the source and target arrays only with valid edges whose source and target nodes are both in the grid. Sinks, outlets and invalid nodes no longer get an outgoing edge pointing to the fictitious -1 node.

flow_routing_d8_carve now returns a topologically sorted array of /nodes/ and a grid of flow directions. This behaves identically to how it did before, but some things have been renamed and comments updated to reflect this new interpretation of the output. There is a node for every pixel in the grid, and they are sorted according to their position in the topological order of the flow network.

flow_routing_d8_targets appends edges to the source and target arrays. These must still be as large as the DEM, because the number of edges is unknown, but flow_routing_d8_targets returns the number of edges, and the source and target arrays can be resized in the host language if needed. An alternative would be to enumerate edges during the flow routing and return an edge count from
flow_routing_d8_carve. Then one could allocate source and target arrays of the correct size.

Because flow_routing_d8_targets now returns only edges that are within the flow network, we do not need to check that the target node is greater than zero in flow_accumulation and drainagebasins. The tests in flow_accumulation to ensure that src and tgt are valid are not required /assuming the source and target arrays have been properly validated/. It is the caller's responsibility to provide valid edge lists.

test/random_dem.cpp required significant refactoring to account for this change in the API. Many of the test functions directly manipulate the edge lists, and those had to be modified.:

streamnetwork generates the stream edge list from the flow direction and accumulation data. It now scans the topologically sorted list of nodes to find nodes whose flow accumulation is greater than the threshold. Then it scan through the flow network edges to find edges that are part of the stream network and appends them to a list. We also no longer have to reverse the list of stream edges, so we can get rid of the intermediate stream_edges array.

The check in test_drainagebasins that the downstream target of each edge is greater than or equal to zero is turned into an assertion, because we should not have any of these placeholder edges remaining in source and target.

test_stream_distance required minor changes such as the addition of the edge_count argument. It should work for either a flow network or a stream network, as long as the appropriate edge_count is passed. We use the flow network in the test, so it computes the upstream distance for every pixel in the DEM and then compares it to the upstream distance computed by streamquad_trapz only for pixels in the stream network.