More stack-related performance improvements

[neoncube2]

Building off of #2835, this PR replaces the remaining instances of stack() with stack(0), so that we avoid getting context information for the first frame of the stack, which is expensive.

I also discovered that we were calling traceback.format_list(traceback.extract_stack()[-10:]), which does the following:

1. Extract the entire stack
2. Format the entire stack into strings
3. Throw away everything except for the last 10 formatted frames of the stack

To fix this, I've made extract_stack() use the limit=10 parameter.

Finally, I optimized _path_utils.py to only retrieve two frames from the stack, instead of retrieving the entire stack and then discarding everything but the second frame :)

[neoncube2]

@mxschmitt Sorry, I accidentally closed and reopened the pull request '-_-

[neoncube2]

@mxschmitt Woohoo, thanks! ^_^