Internal fido2.hid API is racy

[riastradh]

The mechanism of using list_descriptors and then open_descriptor as implemented leads to the following sequence of actions:

1. open /dev/whateverN
2. probe with ioctls or similar
3. close /dev/whateverN
4. put the results in a list
5. pick one
6. reopen /dev/whateverN
7. do FIDO HID on it

At this point, the identity of /dev/whateverN may have changed, if the port where HID N was connected from is disconnected and a new device inserted in its place. If that happens, fido2.hid will happily try to use the new device as if it were the old device, even if it bears no resemblance to the old one -- it may not be a FIDO HID at all.

To avoid this race, it is necessary to avoid the close/reopen cycle, or at least to verify nothing about the descriptor (on which any decisions were made, anyway) has changed since the close/reopen cycle.