There are two sources for examples that go beyond the cases you listed.
The first source is the theory of geodesic laminations on surfaces (I would look this up not just on wikipedia but in the original sources by W. Thurston, and/or the book by Casson and Bleiler). In this theory are lots of new types of geodesics for which the parameterizing map --- which is $\mathbb R \mapsto S$ for a non-closed leaf and $S^1 \mapsto S$ for a closed leaf --- is one-to-one, also known as simple geodesics. If the resulting subset of $S$ is denoted $\lambda$, and if I let $\Lambda$ denote the the closure of $\lambda$, then $\Lambda$ will always have measure zero, so it is never equal to $S$ (this was known by Nielsen). Also, $\Lambda$ will have a unique decomposition as a union of simple geodesics. This kind of subset $\Lambda$ is called a geodesic lamination on $S$. One way to construct interesting geodesic laminations is by taking a sequence
of closed geodesics whose lengths go to infinity, and then taking any limit point of this sequence in the space of closed subsets of $S$ with the Hausdorff topology. If you choose $f : S \to S$ to be a homeomorphism whose mapping class is infinite order and irreducible, and if you define $c_i$ as the closed geodesic homotopic to $f^i(c_0)$, then every leaf of the limit $\Lambda$ will be a non-closed leaf that fails to satisfy your four conditions.
Another source is the theory of Markov partitions. The geodesic flow on $T^1 S$ has a Markov partition; this was known to E. Hopf. The Markov partition is encoded by a finite directed graph $\Gamma$; the flow lines (i.e. the geodesics on $S$) correspond one-to-finite with the bi-infinite directed paths in $\Gamma$. If you now take a proper subgraph of $\Gamma$ which is strongly connected and is not just a circle, and then you take any bi-infinite directed path in that subgraph which does not cycle around a periodic closed path in either direction, then the corresponding infinite geodesic fails to satisfy your four conditions.