Padding machines can cause invalid cell on client intro circs
While stress testing vanguards, I got a dropped cell on a CIRCUIT_PADDING circuit that was previously HS_CLIENT_INTRO:HSCI_INTRO_SENT.
This could be an instance of #31002 (closed). It could be that we are switching to padding, and then some issue happened where the padding machine did not shut down properly. Or it could be another instance of #29927, but with padding involved this time.
Unfortunately I do not have client logs of this to find out, as it was a Tor Browser client. I am working to reproduce elsewhere, but it is rare.
FWIW, this is harmless. I will be adding an exemption to vanguards to demote the log to INFO there, but still close the circuit (since it is padding anyway and not used anymore).