style: cargo fmt --workspace (whitespace/wrapping only, no semantic change)

Whole-workspace rustfmt pass picked up while iterating on Mamba GPU
backward work. Verified formatting-only via diff sampling; no logic
changed.

Co-Authored-By: Claude Sonnet 5 <[email protected]>

crates/production/rtx-inference/src/lookahead.rs

@@ -284,7 +284,12 @@ impl LookaheadDecoder {
     ///
     /// # Returns
     /// The newly generated tokens (does not include `initial_tokens`).
-    pub fn decode<F>(&mut self, initial_tokens: &[u32], max_new_tokens: usize, score_fn: F) -> Vec<u32>
+    pub fn decode<F>(
+        &mut self,
+        initial_tokens: &[u32],
+        max_new_tokens: usize,
+        score_fn: F,
+    ) -> Vec<u32>
     where
         F: Fn(&[u32]) -> u32,
     {
@@ -313,7 +318,11 @@ impl LookaheadDecoder {
                         // The prefix spans the end of tokens and the start of draft.
                         let from_tokens = prefix_len - draft.len();
                         let token_start = tokens.len().saturating_sub(from_tokens);
-                        tokens[token_start..].iter().chain(draft.iter()).cloned().collect()
+                        tokens[token_start..]
+                            .iter()
+                            .chain(draft.iter())
+                            .cloned()
+                            .collect()
                     }
                 } else {
                     // Not enough history yet — build what we can.
@@ -380,9 +389,7 @@ impl LookaheadDecoder {
             // ----------------------------------------------------------
             // 4. Extend sequence
             // ----------------------------------------------------------
-            let to_add = accepted_this_step
-                .len()
-                .min(remaining);
+            let to_add = accepted_this_step.len().min(remaining);
             let accepted_slice = &accepted_this_step[..to_add];
 
             tokens.extend_from_slice(accepted_slice);
@@ -399,8 +406,8 @@ impl LookaheadDecoder {
         }
 
         // Update derived statistic.
-        self.stats.avg_tokens_per_step = self.stats.total_tokens_generated as f32
-            / self.stats.total_steps.max(1) as f32;
+        self.stats.avg_tokens_per_step =
+            self.stats.total_tokens_generated as f32 / self.stats.total_steps.max(1) as f32;
 
         generated
     }
@@ -466,7 +473,10 @@ mod tests {
     fn test_ngram_cache_empty_prefix_returns_empty() {
         let cache = NGramCache::new(3, 100);
         let candidates = cache.top_candidates(&[99, 100], 5);
-        assert!(candidates.is_empty(), "unknown prefix must return empty vec");
+        assert!(
+            candidates.is_empty(),
+            "unknown prefix must return empty vec"
+        );
     }
 
     #[test]
@@ -518,7 +528,10 @@ mod tests {
         decoder.warm_cache(&[10, 20, 30, 40, 50]);
         // After warming, the prefix [10,20] should yield at least one candidate.
         let candidates = decoder.cache().top_candidates(&[10, 20], 3);
-        assert!(!candidates.is_empty(), "warm_cache should populate the cache");
+        assert!(
+            !candidates.is_empty(),
+            "warm_cache should populate the cache"
+        );
     }
 
     // ------------------------------------------------------------------
@@ -707,6 +720,9 @@ mod tests {
     fn test_decode_zero_new_tokens_returns_empty() {
         let mut decoder = LookaheadDecoder::new(LookaheadConfig::default());
         let generated = decoder.decode(&[1, 2, 3], 0, incrementer);
-        assert!(generated.is_empty(), "max_new_tokens=0 should return empty vec");
+        assert!(
+            generated.is_empty(),
+            "max_new_tokens=0 should return empty vec"
+        );
     }
 }