While that script is better than the manual approach (which is what I ended up doing for my JLCPCB order, after trying a bunch of things), I don’t believe that package-name-based adjustment is the fully-correct general solution.
Look at my previous post referencing the EIA-481-D standard, IEC61188 standard, and IPC-7x51 standards. With EIA the rotation correction depends on the part function, not just the part shape.
IF the ad hoc script does become the common approach, it’s going to need quite a bit more work extending the tiny correction list. And even with that database, it’s eventually going to need a way to move per-component rotation correction from the schematic into pcbnew, preferably in a way that changing the footprint is likely to work correctly.
On a related point, this does make it obvious that the default footprint orientation indications aren’t especially clear.