In that case, 12 to 37V would be good - that leaves you with better than 30V Vak with the cathode bias at 7V, which should be enough to get good cut-off even with the max 4.5V (+/- 2.25V) filament drive. On the low end, with the cathode bias pot all the way up, Vak will only be about 5V - useful for those who want to experiment how low they can go and still get a VFD to illuminate.
I'd suggest presets on the board with unpoplulated footprints for fixed resistors in parallel + a footprint bridged by a thin track connecting the wiper, so once the required voltage asre finalised, the user can fit fixed resistors for the desired divider ratio, and either disconnect the preset, or can limit the preset's adjustment range, by putting a relatively large fixed resistor in series with its wiper.
I more or less described the whole variable cathode bias circuit - the only thing I didn't mention was the emitter goes to the C.T of the two resistors instead of the original Zener cathode. Circuit operation: the 100K preset + 100K resistor above it form a potential divider providing a bias voltage between 0V and 1/2 HT, while drawing under 0.2mA from the HT rail, and the PNP Darlington decreases the impedace of the bias voltage by its h
FE, but also adds two Vbe drops to the bias voltage. Assuming a minimum h
FE of 1000, that means the equivalent impedance of the bias circuit, as seen at the Darlington emitter will always be under 100R and typically will be half that or less.
You may also wish to consider doing a Microchip (Supertex)
HV5812 breakout board. Its a dumb SPI VFD anode & grid driver with blanking and 20 outputs that can do up to 80V HT and source up to 25mA, is still in production and is reasonably affordable at around $2 each. Do four daisy-chained boards in a strip and let the user cut them up and populate the SPI headers according to the number of outputs they need - one HV5812 will handle most small VFDs, up to 7 segments, 10 digits + commas annunciators etc, two will handle virtually all starburst displays , and three will handle up to 24 characters of 5x7 dot matrix + cursors. You still need a MCU to handle the multiplexing, but as its SPI clock is up to 5MHz, there's no problem running it at fairly high refresh rates.