Over the years we've had many, many issues with GTK+ and Python
disagreeing about what constitutes a line break. The actual bug that has
prompted this rewrite is just another one in the long series of wild
issues, but it's *so annoying* that rather than work through the insane
logic I wrote before, I've just changed the whole approach and made our
treat-my-textbuffer-as-a-list-of-lines shim use GTK+ APIs to do its line
calculation logic.

I'm was worried that this will be slow - it is after all a _lot_ more
function calls and we're crossing through GObject introspection a bunch,
but it... seems okay?

One of the minor weird side-effects here is that this accessor
absolutely enforces that regex filters only apply to a single line. I
considered options for maintaining our very-broken status quo, but they
were very, very difficult by comparison, and any multi-line filters
are completely unsupported.