Robust title derivation for Subtext content

[cdata]

As of #51 , we will have basic sphere-to-HTML generation landed. In the course of that change, it became apparent that we will need a good, deterministic strategy for deriving titles from Subtext content. We currently use a title header if present, but otherwise we assume there is no title. We could also look at the content itself to attempt to derive the title (e.g., from a heading or a first sentence). We could also use a slug part of a slashlink and de-slugify it. Maybe even subconsciousnetwork/subtext#19 can offer us some options.