Implement parallel image processing with rayon thread pool

The executor now uses rayon's thread pool for parallel processing when
thread_count > 1. Progress updates are sent via mpsc channel from worker
threads. Falls back to sequential processing for thread_count = 1.

pixstrip-core/src/executor.rs

@@ -1,7 +1,9 @@
-use std::sync::atomic::{AtomicBool, Ordering};
-use std::sync::Arc;
+use std::sync::atomic::{AtomicBool, AtomicU64, AtomicUsize, Ordering};
+use std::sync::{Arc, Mutex};
 use std::time::Instant;
 
+use rayon::prelude::*;
+
 use crate::encoder::OutputEncoder;
 use crate::error::{PixstripError, Result};
 use crate::loader::ImageLoader;
@@ -77,6 +79,155 @@ impl PipelineExecutor {
     }
 
     pub fn execute<F>(&self, job: &ProcessingJob, mut on_progress: F) -> Result<BatchResult>
+    where
+        F: FnMut(ProgressUpdate) + Send,
+    {
+        let start = Instant::now();
+        let total = job.sources.len();
+
+        if total == 0 {
+            return Ok(BatchResult {
+                total: 0,
+                succeeded: 0,
+                failed: 0,
+                cancelled: false,
+                total_input_bytes: 0,
+                total_output_bytes: 0,
+                errors: Vec::new(),
+                elapsed_ms: 0,
+            });
+        }
+
+        // Use single-threaded path when thread_count is 1
+        if self.thread_count <= 1 {
+            return self.execute_sequential(job, on_progress);
+        }
+
+        // Build a scoped rayon thread pool with the configured count
+        let pool = rayon::ThreadPoolBuilder::new()
+            .num_threads(self.thread_count)
+            .build()
+            .map_err(|e| PixstripError::Processing {
+                operation: "thread_pool".into(),
+                reason: e.to_string(),
+            })?;
+
+        let completed = AtomicUsize::new(0);
+        let succeeded_count = AtomicUsize::new(0);
+        let failed_count = AtomicUsize::new(0);
+        let input_bytes = AtomicU64::new(0);
+        let output_bytes = AtomicU64::new(0);
+        let cancelled = AtomicBool::new(false);
+        let errors: Mutex<Vec<(String, String)>> = Mutex::new(Vec::new());
+
+        // Channel for progress updates from worker threads to the callback
+        let (tx, rx) = std::sync::mpsc::channel::<ProgressUpdate>();
+
+        let cancel_flag = &self.cancel_flag;
+        let pause_flag = &self.pause_flag;
+        let pause_on_error = self.pause_on_error;
+
+        // Run the parallel work in a scoped thread so we can drain progress concurrently
+        std::thread::scope(|scope| {
+            // Spawn the rayon pool work
+            let tx_clone = tx.clone();
+            let completed_ref = &completed;
+            let succeeded_ref = &succeeded_count;
+            let failed_ref = &failed_count;
+            let input_bytes_ref = &input_bytes;
+            let output_bytes_ref = &output_bytes;
+            let cancelled_ref = &cancelled;
+            let errors_ref = &errors;
+
+            let worker = scope.spawn(move || {
+                pool.install(|| {
+                    job.sources.par_iter().for_each(|source| {
+                        // Check cancel
+                        if cancel_flag.load(Ordering::Relaxed) {
+                            cancelled_ref.store(true, Ordering::Relaxed);
+                            return;
+                        }
+
+                        // Wait while paused
+                        while pause_flag.load(Ordering::Relaxed) {
+                            std::thread::sleep(std::time::Duration::from_millis(100));
+                            if cancel_flag.load(Ordering::Relaxed) {
+                                cancelled_ref.store(true, Ordering::Relaxed);
+                                return;
+                            }
+                        }
+
+                        if cancelled_ref.load(Ordering::Relaxed) {
+                            return;
+                        }
+
+                        let file_name = source
+                            .path
+                            .file_name()
+                            .and_then(|n| n.to_str())
+                            .unwrap_or("unknown")
+                            .to_string();
+
+                        let current = completed_ref.fetch_add(1, Ordering::Relaxed) + 1;
+
+                        let _ = tx_clone.send(ProgressUpdate {
+                            current,
+                            total,
+                            current_file: file_name.clone(),
+                            succeeded_so_far: succeeded_ref.load(Ordering::Relaxed),
+                            failed_so_far: failed_ref.load(Ordering::Relaxed),
+                        });
+
+                        let loader = ImageLoader::new();
+                        let encoder = OutputEncoder::new();
+
+                        match Self::process_single_static(job, source, &loader, &encoder) {
+                            Ok((in_size, out_size)) => {
+                                succeeded_ref.fetch_add(1, Ordering::Relaxed);
+                                input_bytes_ref.fetch_add(in_size, Ordering::Relaxed);
+                                output_bytes_ref.fetch_add(out_size, Ordering::Relaxed);
+                            }
+                            Err(e) => {
+                                failed_ref.fetch_add(1, Ordering::Relaxed);
+                                if let Ok(mut errs) = errors_ref.lock() {
+                                    errs.push((file_name, e.to_string()));
+                                }
+                                if pause_on_error {
+                                    pause_flag.store(true, Ordering::Relaxed);
+                                }
+                            }
+                        }
+                    });
+                });
+                // Drop sender so the receiver loop ends
+                drop(tx_clone);
+            });
+
+            // Drop the original sender so only the worker's clone keeps the channel open
+            drop(tx);
+
+            // Drain progress updates on this thread (the calling thread)
+            for update in rx {
+                on_progress(update);
+            }
+
+            // Wait for worker to finish (it already has by the time rx is drained)
+            let _ = worker.join();
+        });
+
+        Ok(BatchResult {
+            total,
+            succeeded: succeeded_count.load(Ordering::Relaxed),
+            failed: failed_count.load(Ordering::Relaxed),
+            cancelled: cancelled.load(Ordering::Relaxed),
+            total_input_bytes: input_bytes.load(Ordering::Relaxed),
+            total_output_bytes: output_bytes.load(Ordering::Relaxed),
+            errors: errors.into_inner().unwrap_or_default(),
+            elapsed_ms: start.elapsed().as_millis() as u64,
+        })
+    }
+
+    fn execute_sequential<F>(&self, job: &ProcessingJob, mut on_progress: F) -> Result<BatchResult>
     where
         F: FnMut(ProgressUpdate) + Send,
     {
@@ -102,7 +253,6 @@ impl PipelineExecutor {
                 break;
             }
 
-            // Wait while paused (check every 100ms)
             while self.pause_flag.load(Ordering::Relaxed) {
                 std::thread::sleep(std::time::Duration::from_millis(100));
                 if self.cancel_flag.load(Ordering::Relaxed) {
@@ -129,7 +279,7 @@ impl PipelineExecutor {
                 failed_so_far: result.failed,
             });
 
-            match self.process_single(job, source, &loader, &encoder) {
+            match Self::process_single_static(job, source, &loader, &encoder) {
                 Ok((input_size, output_size)) => {
                     result.succeeded += 1;
                     result.total_input_bytes += input_size;
@@ -149,8 +299,7 @@ impl PipelineExecutor {
         Ok(result)
     }
 
-    fn process_single(
-        &self,
+    fn process_single_static(
         job: &ProcessingJob,
         source: &crate::types::ImageSource,
         loader: &ImageLoader,