tinyusb: correct SAMD51 serial number extraction

[kaysievers]

TinyUSB does not match the exported serial numbers of the classic USB
stack.

The current code serializes the 32-bit words backwards, and the byte
nibbles are also backwards serialized.

A SAMD51 is exported like:
EFADAF3113347335020202938343E0FF

while it should read:
13FADAFE5337433139202020FF0E3438

[hathach]

Great thanks again. I am not aware of this backward serial. Copy this serial extraction code from circuitpython :D

[dhalbert]

We can fix this in CircuitPython too. I think we were not aware there was a canonical ordering used by others: we were more interested in it just being unique.

[kaysievers]

Yeah, I don't think there is a strict rule or defined order, it's just bits and using all of them should be universally unique. It's just more common though to start the serialization of bytes with the first word and not the last.

With this change, the USB mass storage bootloader, the Arduino classic USB, the tinyUSB stack all show the same serial number for the same hardware, that's why we synchronized it.

[hathach]

Oh, I forgot that we need to stay consistent with circuitpython. @dhalbert Let's me know your decision, if you decide not to change cpy, we will probably revert this PR. Although it is not the same as Arduino stack, it is more important to stay consisent with circuitpython

[kaysievers]

Note, that the code also serialized the individual bytes backwards. 0x0a showed up as "A0". Did python really do the same?

[kaysievers]

Although it is not the same as Arduino stack, it is more important to stay consisent with circuitpython

If all goes well with my tinyusb tests, I would love to have a new firmware version for my MIDI devices switching to it. But if the serial numbers change with with the switch from the Arduino version, they will no longer be recognized as the same devices by the operating system they are connected to, and all the config in the music apps needs to be redone.

[dhalbert]

@kaysievers The CircuitPython code was copied into this implementation before this fix, so yes, it did.
@hathach I filed an issue in CircuitPython, and can fix it sooner rather than later.