Also record the raw device events through BPF

[bentiss]

There are multiple case where the events forwarded to hidraw are not the pristine ones from the device:

• if a HID-BPF filter is inserted and changes the incoming data
• if a driver changes the data in its .raw_event()

Let's inject a new HID-BPF filter in front of anyone else, and use it to retrieve the data from the device.

This is controlled by a new --bpf flag or will also be enabl;ed if we detect that a HID-BPF filter is in used on the given device.

Further improvements can be done:

• ~support more than 64 bytes of data by injecting more than one element into the ringbuf~ done
• ~detect cases where the chances of the event stream not to be tempered by the kernel (no HID-BPF filters and hid-generic in use or hid-multitouch)~

[bentiss]

And I've now fixed the github workflow.

So I think we can have another look at this now.

Note: I've also added a commit to support packets longer than 64 bytes, but the syntax is not very clean as there will be 2 B: tags: one for the beginning with the timestamp and the total size, and all the rest will have just B: and the remaining bytes.

[whot]

Let's get this merged and fix up any remaining things later, thanks for the patches!