This adds an experimental library API for searchers: depth-first search and breadth-first search.

The current API proposal I ended up with is as follows

// created a new search context that can be re-used across multiple searches
tinygraph_dfs_s ctx = tinygraph_dfs_construct(graph);

// for this search start at node 0
tinygraph_dfs_set_start(ctx, 0);

// step through the search space one by one
while (!tinygraph_dfs_is_done(ctx)) {
    uint32_t next = tinygraph_dfs_step(ctx);
    ..
}

// to re-use the same context, clear it and then step through another search
tinygraph_dfs_clear(ctx);

The main idea is that the search context can keep internal state (stack or queue and a bitset) and cache their construction and allocation between multiple searches. For example for all subsequent searches after the very first one we can simply re-use the allocated bitset and can re-use the stack or queue's internal allocations.

Open questions

• Should clear and set_start be two separate functions or should it just be one function that automatically clears the internal state whenever a new start node is set? What are the pros and cons of this design decision?
• Should we provide more customization at this point already e.g. a max depth at which to stop the search?
• Should the search stepping only return newly discovered nodes or rather return (s, t) node tuples so that the user has more information than just the new node?
• How would an easy interface look like so that we can wrap it e.g. into a Python iterator interface for ease of use?

Once we get this API into the standard libtinygraph.h header we better have a solid interface figured out. The alternative is adding a separate experimental header just for searches.