WIP: Replace python-based JS extraction script with run_shell

[TLATER]

This is an initial stab at #270 and removes the python dependency in favor of a heavier reliance the shell.

While not ideal right now, a resolution of bazelbuild/bazel#5511 would allow us to do the file grepping in starlark

We could use entirely native rules this way, only calling out to the jar binary directly, which should be included with any jdk, and therefore result in a rule that is much leaner on dependencies while also more portable.

Note: This hasn't been tested yet. I've thought a few times I did using the JS example, but that still builds fine if I remove all relevant code, so I don't think the modified rule actually ran ;) This is a start, and an approach, but not a ready patch. I've run out of time for this for now, so I'm pushing it for future reference.

This also still crucially lacks an implementation for .js.map files, which is technically trivial, but I'd like to ideally do it without invoking more shells.