A blank line will not appear in ``lnotab`` as returned by
``_find_executable_linenos``. Alternately, ``lnotab`` might be set to an
empty dict. Either way, the check for whether a line is blank can be
done after checking for whether the ``lineno`` is in ``lnotab``.

There's no need to check for an empty line. Either we hit the line and
wrote the count, we didn't hit the line and wrote an annotation, or we
want to print some spaces to indent the line.

The lnotab parameter shouldn't be renamed, for compatibility, but it
could use some further explanation. One natural reading of it would be
"lines without tabs", but it's really a "line number table" -- a mapping
of line numbers to possible executable lines.

One nuance to that "table" is that ``_find_executable_linenos`` will
ignore some code that the user might want to highlight as unreachable.
For example, code inside a condition like ``if False:`` will be
excluded.

Further, it appears ``lnotab`` doesn't need to be a dict. A set would
suffice and might reduce the memory requirements of the trace module. If
the ``_find_lines`` function is used only by the trace module, as the
underscore prefix suggests, it could be refactored to return a set of
line numbers instead of a mapping of line numbers to lines.

Whether the show_missing parameter is true or false, the cover file
should be written. The ``lnotab`` parameter will be used simply to
decide whether to annotate unreached lines with some extra characters.