Optimize isr handler once more by removing gpio -> index conversion at runtime
Drop atomics for now, since we can also make the whole operation atomic
by locking GPIO interrupts while reading and setting the counter value
And we don't have another thread / core to worry about