Fix missed Xteink navigation during rapid taps and redraws

[LittleBidet]

Quick Xteink X3/X4 navigation taps can be missed for two reasons: the board handler stopped sampling for 200 ms after any press, and the shared input task wiped every unread flag after about 75 ms, even while loopTask was blocked in an e-paper panel refresh. A tap landing in that window was raised, wiped, and never seen — the "second press does nothing" report on the X3.

Board handler (boards/xteink-x3-x4/interface.cpp). Read both ladders and the power button on every poll. A rung counts once two consecutive samples agree: releasing Right or Down sweeps the ADC node up through the other bands on its way to idle, and a single transient sample must not fire a stray Prev/Sel/Esc/Up, nor end a hold. A stable idle rearms the ladder; a stable different button is a new press. Each ladder repeats independently at the existing 200 ms interval; LongPress bypasses the gate as before. Costs one 10 ms sample of latency on the first press.

Shared input task (src/main.cpp, E_PAPER_DISPLAY only). While any of the six button flags is unread and LongPress is not set, skip the per-cycle wipe of those flags; touch and keystroke state are still cleared each cycle. Pending is judged on the flags themselves, not AnyKeyPress, because check() drops AnyKeyPress on the first flag it consumes and the others have to survive that. Only one unread event is held at a time: the flags carry no order, so a press stacked on an unread one would be applied in whatever order the UI polls them (move-then-confirm when the user did confirm-then-move). While something is pending, whatever a poll newly raises is cleared, as every poll did before; a flag the handler cleared is never restored, so a press check() consumed meanwhile cannot be put back. Bounded at 1500 ms so an event nobody reads cannot ghost into a later screen. Handlers that assign false while idle (hal_buttons_poll_3, T-Deck Pro) still drop their flags, so those boards keep their previous behaviour. Not while LongPress: display.cpp, mykeyboard.cpp and launcherSelectHeld() poll the raw flags and need them to drop on release. LCD boards are unchanged.

Main menu. On e-paper boards, consume Back/Escape as a no-op on the main menu. Otherwise its unread flag occupies the retained event slot and discards navigation or Confirm presses for up to 1500 ms.

Board handler, measured. A hold repeats only after 400 ms (then every 200 ms), and only for Left/Right/Up/Down. Confirm and Back fire once per press; their 200 ms repeat used to be wiped before the UI saw it, but with retention a half-second Confirm produced a second Sel that landed in the menu the first one opened. Ordinary taps measured up to ~240 ms, so a 200 ms first repeat double-stepped 2 of 20 taps. Main menu consumes an unread Escape so it cannot block navigation for the retention window.

Validation. pio run -e xteink-x3-x4 passes (application image 1,376,688 bytes); seeedstudio-reterminal-e1001 and lilygo-t-deck-pro also build. A temporary host harness mirroring the task-loop block covered retention across 75 ms polls, a second flag surviving the first check(), confirm-then-move dropping the move, a check() racing InputHandler() not restoring the consumed press, the 1500 ms cap, LongPress live state, idle-assigning handlers, and timer wrap.

On an X3 (UC8253 panel), with a temporary trace printing both ladder ADCs on every 10 ms poll and every raised flag, over two runs and 112 press runs: correct button on every press, 0 stray flags on release, 0 release sweeps with an intermediate band held for two consecutive samples (one Right release showed a single Back-band sample at 3854, rejected as intended), hold-Right repeats at 400 ms then 200–240 ms with a clean release and no trailing event, a page tap registered once while Right kept repeating, a half-second Confirm produced exactly one Sel, 21 taps at normal pace produced 21 events. Band centres on this unit: Back 3564, Confirm 2727, Left 1500, Right 0; page 2267 and 5. A press landing during the new screen's first refresh is applied to that screen. git diff --check passes. No test framework is added.

Note for a separate change: this env does not define ARDUINO_USB_CDC_ON_BOOT, so Serial is UART0 on GPIO20/21 (I2C SDA and panel CS here) and the console is unreachable over the X3's USB port. The trace above was captured with that flag set in the environment, not in the repo.

Draft. Remaining UI-level checks are listed in the review thread. Known limits: a second press while one is unread is dropped (same as before this PR); multiple taps on one direction during a refresh coalesce; taps that fall entirely between polls, refreshes longer than 1500 ms, and the explicit resetGlobals() after some redraws can still discard navigation.