Add flushing PIO data

[Problem]
When providing data fast enough, multiple frames could be concatenated
since we only waited 60 microseconds after the last bytes were pushed
to the PIO, not since the last bytes were actually pushed to the
neopixels.

[Solution]
"Flush" the PIO after pushing the last bytes, by waiting until it's
empty.

neopixel-server/src/bin/tcp.rs

@@ -151,6 +151,8 @@ async fn main(spawner: Spawner) {
                         .for_each(|(r, g, b)| ws2812.write(r, g, b));
 
                     num_bytes = None;
+                    // wait for the state machine to write all bytes
+                    ws2812.flush();
                     // let the neopixels latch on
                     Timer::after_micros(60).await;
                 }

neopixel-server/src/bin/udp.rs

@@ -140,6 +140,8 @@ async fn main(spawner: Spawner) {
                         .for_each(|(r, g, b)| ws2812.write(r, g, b));
 
                     num_bytes = None;
+                    // wait for the state machine to write all bytes
+                    ws2812.flush();
                     // let the neopixels latch on
                     Timer::after_micros(60).await;
                 }

neopixel-server/src/ws2812.rs

@@ -91,6 +91,15 @@ impl<'d, P: Instance, const S: usize> Ws2812<'d, P, S> {
         }
     }
 
+    // see https://github.com/rp-rs/ws2812-pio-rs/blob/944494ca9dad73933f700408c3054c8f14c78998/src/lib.rs#L262-L263
+    pub fn flush(&mut self) {
+        // clear stalled flag first
+        self.sm.tx().stalled();
+        while !self.sm.tx().empty() && !self.sm.tx().stalled() {
+            cortex_m::asm::nop();
+        }
+    }
+
     // left here for future experiments
     pub async fn _write_dma(&mut self, data: &[u32]) {
         self.sm.tx().dma_push(self.dma.reborrow(), data).await;