[jeffreym-emt]

Closes #5326

v4l2_scan_devs() releases dev_pool during refresh but leaves dev_info pointing to freed memory. When the device count stays the same or decreases, allocation can be skipped, causing writes to freed memory and subsequent crashes during format enumeration.

This MR:

• Resets dev_info to NULL after releasing its pool, ensuring a fresh device array is allocated.
• Replaces the ineffective j < fmt_cnt check with pdi->info.fmt_cnt == 0, preventing devices without usable formats from being advertised.

Tested in the affected application and confirmed that the changes prevent the previously observed crash.

[CLAassistant]

All committers have signed the CLA.

[nanangizz]

Thanks, both changes look right. Clearing dev_info after the pool release makes the allocation unconditional, so a refresh with the same or fewer devices no longer copies into the released pool. And pdi->info.fmt_cnt == 0 replaces a check that could never be true (j == fmt_cnt after the size loop), so a device without any usable size is no longer listed with an empty format list.

Optional: with dev_info always NULL at that point, f->dev_count > old_count || f->dev_info == NULL is always true, so the condition and old_count could be dropped and the array allocated unconditionally.