August 2024

Until now, "arcs" means a complete set of predicted and measured (from, to)
pairs of line numbers.  Branches were determined by finding "from" lines that
appeared in more than one predicted arc.

That scheme found branches that were not true branches, such as the lines in
finally clauses that could jump to more than one place based on how the finally
clause was reached.

Now we are shifting to true branches. To do this, we are removing code that
predicted arcs that aren't part of true branches.  The ideal goal would be to
only predict arcs that are part of branches, but a minimal goal is to stop
predicting arcs that led to false branches.  ie, it's ok to predict an arc if
the arc is the only arc for a given "from" line. Those arcs will be discarded
and won't lead to false branches.

There are many tests that look odd now, because they were testing arc
determination, but they have no branches.  Or the interesting part of the tests
were non-branch arcs, so they aren't visible in the tests anymore.

parser.py likely is working harder than it needs to, since we don't need to find
all arcs.  The new code.co_branches() function might be good enough to replace
it.