fix(layer): debounce the bi-directional switch so one flick moves one layer

[DriftKingTW]

Problem

Flicking the bi-directional switch to change layer sometimes advanced two layers at once.

The handler waited for the pin to read inactive exactly once, with no debounce, so it returned mid-bounce on release:

while (digitalRead(BD_SW_CW) == ACTIVE) {   // exits on the FIRST inactive sample
    delay(10);
}

loop() comes back around in roughly 1 ms — well inside a mechanical switch's 1–10 ms release chatter — so the still-bouncing contact was read as a brand-new flick. The pins are bare INPUT_PULLUP with no hardware debounce.

It only showed up on the switch because switchLayout(int) has no guard time, unlike the no-arg switchLayout(), whose trailing delay(300) masked the problem on the FN-key path.

Fix

• readSwitchPressed() — re-samples the press edge after a short settling delay, rejecting contact chatter and electrical glitches.
• waitForSwitchRelease() — blocks until the contact has read inactive continuously for SW_DEBOUNCE_MS (25 ms), restarting the quiet period on any activity.
• All three switch directions (CW / CCW / push) routed through both.

Cost is nil while the switch is idle: readSwitchPressed() returns on its first digitalRead, so the settling delay only runs on an actual press.

Also guards switchLayout(int) against layoutLength == 0, where wrapping stored layoutLength - 1 == -1 as 255 and indexed past the end of keyConfig.

Testing

• pio run builds clean; no new warnings from src/.
• Flashed to hardware and confirmed by the author: one flick now moves exactly one layer.