If for any reason you can't use a mechanical solution, the TMC2209 datasheet at the bottom of page 11 explicitly notes that the motors acting as generators can back-drive VS.
Attention
Ensure sufficient capacity on VS to limit supply ripple, and to keep power slopes below 1V/µs. Failure to do so could result in destructive currents via the charge pump capacitor. Provide overvoltage protection in case the motor could be manually turned at a high velocity, or in case the driver could become cut off from the main supply capacitors. Significant energy can be fed back from motor coils to the power supply in the event of quick deceleration, or when the driver becomes disabled.
It does not set limits on how long for or how high a current is tolerable, but as this current will be carried by the output stage MOSFET body diodes, I would expect it to tolerate close to its continuous RMS output current rating of 1.4A without issues. If the lack of an official rating for this situation concerns you, add beefy Schottky clamping diodes to each output to keep them between VS and GND.
You also need some sort of shunt regulator overvoltage limiter to keep VS under the max permissible TMC2209 supply voltage and a power diode in series with VS to prevent it back-feeding power to your MCU. There *MUST* be enough bulk decoupling after the VS blocking diode to keep VS transients under the 1V/us limit mentioned above during worst case unpowered stepper acceleration. Its desirable to let VS rise to above its normal operating voltage for the application to minimize resistance to motion due to the motors acting as generators, which also simplifies the design of the overvoltage limiter circuit.
Its also probably desirable to ensure its disabled when the steppers are generating VS. This could be done by powering VCC_IO from its internal regulator 5VOUT, and providing a pullup resistor to 5VOUT on its ENN input.
You then have the problem of ensuring that it control lines cant back-drive the MCU, and also translating logic levels if using a 3.3V MCU. One approach would be to use MOSFET level shifters as used for I2C bus applications, e.g. the classic Philips AN97055 circuit:
(https://www.eevblog.com/forum/beginners/overvoltage-protection-with-a-twist/?action=dlattach;attach=306236;image)
Pullup resistor values depend on application
but for all except its PDN_UART pin being used bidirectionally it would be preferable to use a dual supply level translator which supports partial power-down mode e.g. 74LVC1G45 or 74LVC2G45 with their EN pin strapped according to the desired direction, and the TMC2209 side powered from 5VOUT. Note that apart from the ENN pullup and possible back-driving from PDN_UART, its input only control pins shouldn't require isolation from a 5V MCU.
For the issue of the manual crank spinning under motor drive, it may be sufficient to simply add a push in to engage dog clutch behind the handwheel with a friction brake in the out position if the shaft carries through the dog clutch.