Fix NPE error caused by the second HID endpoint of HID composite devices

[KenRouKoro]

This PR fixes a NullPointerException in the hid4java data reader and device enumerator threads that occurs when a single USB device enumerates multiple HID interfaces (with identical VID/PID/serial numbers) due to the physical device having a second HID endpoint, which previously prevented the receiver's tracking data from being recognized.

Additionally, it resolves a secondary issue introduced by the initial fix: if the second HID endpoint is not a SlimeVR HID endpoint, it causes a loop of [TrackerServer] Reopening device ${device.serialNumber} after no data received. This happens because the system attempts to handshake with the second HID endpoint as if it were a SlimeVR device, despite there being no corresponding return data. To address this, I have introduced a keepFirstHidPerPhysicalDevice filter to block extra HID endpoints.

[loucass003]

@sctanf can you double check that one 🙏

[sctanf]

The logic was a little hard to follow, but it seems to be okay.
At line 265: I think it could be changed to just remove devices not in keptSet, so making the extra sets enumeratedSet and droppedSet is redundant and probably will make it clearer to understand. Assuming the rest was done for readability which is okay I think

[sctanf]

I am not aware of a use case to have multiple HID interfaces, is this for functionality outside of SlimeVR?