Eye tracker: ft-eyegrab checks only the slot each camera writes next

While copying, ft-eyegrab woke every 300 us (about 1,500 to 3,000 times a second) and
fingerprinted all eight slots each time: 8 x 256 strided reads from DMA-BUF memory.

- Each look now checks only the slot each camera writes next. The order is known (camera 0
  3,0,1,2; camera 1 7,5,4,6,5,7,6,4), and the next slot follows from the last two; the table
  starts from those orders and learns from every frame, so a SteamVR update that changes
  them costs a few seconds of full scans, not frames.
- A camera with nothing in its expected slot 1.5 frames after its last one, or with no
  order yet, gets all four slots checked, as before. A frame that turns up in an unexpected
  slot means full scans for that camera for 2 s.
- A slot's fingerprint is taken again when it stops being one of the two in use, so a later
  check sees only a new frame. A frame is still passed on when its camera starts the frame
  after next.
- Between frames it sleeps until 2.5 ms before the next is due, then looks every 1 ms, with
  0.5 ms of timer slack (PR_SET_TIMERSLACK, --share only). With no frames from either
  camera for 0.5 s (headset off) it looks every 4 ms.
- --rec keeps its 0.3 ms polls (and the expected-slot checks), for its timestamps.

Tested offline by building poll_frames against simulated cameras that write each frame in
four bursts, the last after the next frame starts (6 s, both cameras): 1,076 frames passed
on, none torn or skipped, with the known orders and with camera 1 in a different order.
Wakeups 1,486/s -> 207/s, the poller's CPU 3.6% -> 0.8% of a core (in plain memory; the real
DMA-BUF reads cost more), and a frame's start is seen 1.35 ms after it begins on average
instead of 0.76. With a camera stalling 15 ms every 2 s, the old poller passed on 22 torn
frames and the new one 8 or fewer.

Built (glibc 2.38 symbols at most, the host has 2.39), not installed: it runs as root from
/etc/frametop, so it takes effect only after gaze/tracker/install.sh (sudo).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

gaze/tracker/findings.md

@@ -63,6 +63,13 @@ reading only.
     frame when its camera starts the frame after next (no slot is rewritten sooner than three
     frames), ignores late writes to the frame just finished, and saves from a separate
     thread. fit1 may hold a few percent of torn frames.
+  - `--share` (2026-10-03) checks only the slot each camera writes next, from the two before
+    (the orders above, learned again if they change; all four slots for 2 s after a frame turns
+    up elsewhere), sleeps until 2.5 ms before the next frame is due, then looks every 1 ms with
+    0.5 ms of timer slack. Against the 0.3 ms poll of all eight slots, on simulated cameras:
+    207 wakeups a second instead of 1,486, 0.8% of a core instead of 3.6% (more on the real
+    DMA-BUF memory), no torn or skipped frames, and a frame's start seen 1.35 ms late on
+    average instead of 0.76. `--rec` still polls all slots every 0.3 ms, for its times.
   - Eye tracking stops when the headset is off ("HMD off, stopping eye tracking"), so
     recordings are empty then.
 - **Which camera is which eye** (capture fit1, 2026-09-29, closing one eye at a time):