Key de-bouncing in 4x4 matrix....

Consider separating the scanning from the processing.
So remove your delays from the scanning - the scanning is fine with just enough delays that each signal line has time to stabilize before you read out a value. And for most situations, the signals stabilizes "instantly" compared to the speed of the processor.

Require the keyboard scan to produce the same key press n times in a row - then accept it as a press. Then stop accepting key presses and instead require the keyboard scan to produce "empty" n times in a row. Then back to wait for a new key press.

res = scan_keyboard();
if (waiting_for_key) {
    // Waiting for a distinct key press for n scan iterations.
    if (res == -1) {
        consecutive_samples = 0;
        last_res = -1;
    } else {
        if (last_res != res) {
            consecutive_samples = 1;
            last_res = res;
        } else if (++consecutive_samples >= debounce_limit) {
            waiting_for_key = false;
            return res;  // return the pressed key
        }
    }
} else {
    // Waiting for distinct keyboard idle for n scan iterations.
    if (res == -1) {
        if (++consecutive_samples >= debounce_limit) {
            waiting_for_key = true;
            last_res = -1;
        }
    } else {
        consecutive_samples = 0;
    }
}

// no key pressed
// or multiple keys pressed
// or state not properly debounced yet
return -1;