I can explain that difference:
When the first argument of GeoGraphics or DynamicGeoGraphics contains several point-locations then we show a map of a region that contains all those locations, with a bit (say around 10%) of padding. The same thing happens when the first argument contains something with a finite size (say a disk, or a polygon).
But when there is only one point-location we cannot infer a notion of size, and we default to a radius of 10 miles. This is what happened when you restricted the region to 0.5 km: GeoEntities only found one entity, and therefore GeoGraphics and DynamicGeoGraphics did not know how to compute a map radius. Adding GeoRange -> region or GeoRange -> Quantity[0.5, "Kilometers"] would indicate the scale you want, and solve the issue.
And yes, we don't yet have entities for streets.