Thank you @mcc for this rundown, it helped me try to pinpoint the Windows driver issue/s.
For a long while I thought it was due to library changes in the STM32Cube_FW_F4_V1.25.1 updates. But no, the libs check out alright.
Turns out Windows doesn't like audio devices with zero channels. And that there's some issue with the new audio driver, so that compiling it in MIDI-only mode caused it to hang whenever there was USB activity. And that for some reason Windows doesn't like our driver with output channels configured.
And most annoying of all, because I knew about it but still didn't realise that's why my testing failed: Windows has issues with MIDI devices resetting (as when you've just flashed it) while there's a program connected to it (such as Chrome or Edge). Mostly it works, but sometimes it hangs the browser and does all sorts of weird things. Frequently it looks as if the device is connected, but no data is sent or received. Restarting Chrome doesn't help. Sometimes doing a 'hard' restart works, ie going to Task Manager and killing all the background Chrome processes. Suffice to say, Windows and I don't enjoy each other's company.
Long story short: Magus firmware, develop branch should now work on Windows.
Key configuration settings:
#define AUDIO_CHANNELS 2
#define USE_USBD_AUDIO_TX // microphone
/* #define USE_USBD_AUDIO_RX // speaker */
Upshot is, you've now got a USB Audio interface that gives you the Magus stereo inputs as 48kHz 16-bit audio.