webos: only ask /sys/class/input about input nodes

[mariotaku]

/sys/class/input only lists evdev and js nodes, so wherever /sys/dev is missing (every production jail) the presence probe reported every hidraw node as absent. Other names now fall through to the open() probe. Cherry-picked from 495223e.

Verified: builds with the webOS toolchain (CI config), no warnings.

Untested on a device under a production (native/native_game) jail. The check: under jailer -t native with a hidraw gamepad, unplugging it should be reported; and dead /dev/hidrawN slots should fail open() with ENODEV/ENXIO (EACCES would make them read as present).

🤖 Generated with Claude Code

[mariotaku]

Tested on lk5900 (webOS 4.4.3) under jailer -t native (no /sys/dev), comparing 649b462 (before) with 2dd4597 (after), with a DualShock 4 (054c:09cc) on USB.

Snapshot, pad plugged in: /dev/hidraw0 opens fine in both builds. Before: verdict ABSENT, HIDRAW=0x0. After: verdict PRESENT, HIDRAW=0x1. evdev/js identical.

Snapshot, pad unplugged: every hidraw slot fails with ENODEV and reads ABSENT in both builds, so no phantom devices. The jail's copies are 0660 root:5000 and the app runs as gid 5000, so the driver itself answers. Same result earlier in native_game and native_devmode, and on up7560 (webOS 6.5.2), which has no /dev/hidraw* at all until a device binds.

Live hot-plug (testwebosuevent, 2 plug/unplug cycles each):

	before	after
netlink add/remove for hidraw0	4	4
poll saw hidraw0 change	0 ("the bitmask could express 0")	4 (0.2–1 s after netlink)
transitions the poll cannot see	4	0

[115.198] netlink  add    /dev/hidraw0
[115.846] poll     add    hidraw/0 (netlink was 648ms ahead)
[116.671] netlink  remove /dev/hidraw0
[116.855] poll     remove hidraw/0 (netlink was 184ms ahead)

Only the 3.4.x build ran on the TV; the 2.30.x change has the same logic.

🤖 Generated with Claude Code