* Add safety for empty stack on shutdown.

* Correct line length linting issue.

---------

Co-authored-by: Ned Batchelder <ned@nedbatchelder.com>

* cache the bound method of _trace on self

this speeds up pure python tracing because we don't have to re-create a
bound method object all the time

* optimize checking whether a file should be traced

the optimization works based on the following heuristic: in a
majority of cases, functions call other functions in the same file. In
that situation we don't have to re-check whether we should trace the
file

* fix optimization in the presence of contexts

* fix too long line

use the python >= 3.7 feature of being able to disable line tracing by
setting the frame attribute f_trace_lines to False. This can be used for
the frames of functions that we aren't collecting coverage information
for (eg those functions in the stdlib).

This speeds up the pure python tracer in CPython by ~9x and in PyPy by
80% on a coverage run of one realistic project that I tried.