WebDriver script should rely more on in-page functions

The script in `tools/conformance.py` is very tightly bound to the structure of the conformance test harness, selecting elements by class name and then navigating the DOM. This is fairly brittle, because any change to the harness could break this script.

I suggest that instead the harness expose some JS APIs that the WebDriver can call to perform the operations. They would include:

- a list of tests (with a potential filter for directories/names)
- a way to invoke a particular test
- a signal that the test has passed/failed