Here are some problems you mentioned and potential solutions that could be solved by the community:
Geocoding: I assume you are referring to the user routing to a specific address destination and that address need to be converted to a lat/lon? I think yahoo, Google, and MS virtual earth expose their geocoding APIs. Also there is this SRC USA open source option witch might help for United States users only.
Road Flow concerns: Intersections, traffic Light, Stop Signs, Road barriers, Speed limits, One way streets.
I believe these can all be addressed by lots of probes reporting data and then an algorithm to process that data. Examples:
Does a highway connect with a secondary road? Algorithm asks, has a user id ever turned onto the road, if so where did they do so and how many users have done so? Why is the chance this intersection is routable?
Road speed: Must users won't care what the speed limit is, they can see the sign for that when they get there. The routing algorithm just needs to know the average road speed for the day and time the travel is occurring.
Lights and stop signs: By watching the flow of traffic, the algorithm should be able to determine what traffic control mechanism is in place. Do vehicles always stop? Do they stop for 3 minutes? This would allow light times or stop signs to be calculated into routing, which isn't currently being done with most commercial solutions.
Road barriers and one-way streets: watching the probes also solves these questions.
The algorithms would also be smart enough to possibly throw at anomalies. Let's just say for example that I am probe. I report with my iPhone. I regularly bike the wrong direction on one-way streets and speed 20 miles over the speed limit. The algorithm would eventually throw out certain parts of my data that are way outside the norm and negatively weight all of my other reports.
All of this requires huge numbers of probes and smart people to work on the project. Both of these things exist now and they didn't a few years ago.