style: apply rustfmt across all crates and demos

Consistent formatting pass: line wrapping, import sorting, trailing
whitespace removal, let-chain indentation, merged derive attributes,
and unsafe block reformatting.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

crates/training/rtx-transformers/src/layers/sage_attention.rs

@@ -89,7 +89,7 @@ impl Default for SageAttentionConfig {
 
 impl SageAttentionConfig {
     /// Create config for a specific model size
-    #[must_use] 
+    #[must_use]
     pub fn for_model(num_heads: usize, head_dim: usize) -> Self {
         Self {
             num_heads,
@@ -99,28 +99,28 @@ impl SageAttentionConfig {
     }
 
     /// Enable GQA with specified number of KV heads
-    #[must_use] 
+    #[must_use]
     pub fn with_gqa(mut self, num_kv_heads: usize) -> Self {
         self.num_kv_heads = Some(num_kv_heads);
         self
     }
 
     /// Enable MQA (single KV head)
-    #[must_use] 
+    #[must_use]
     pub fn with_mqa(mut self) -> Self {
         self.num_kv_heads = Some(1);
         self
     }
 
     /// Set causal masking
-    #[must_use] 
+    #[must_use]
     pub fn causal(mut self, causal: bool) -> Self {
         self.causal = causal;
         self
     }
 
     /// Disable smooth quantization for faster but less accurate inference
-    #[must_use] 
+    #[must_use]
     pub fn fast_mode(mut self) -> Self {
         self.use_smooth_quant = false;
         self.per_token_quant = false;
@@ -141,7 +141,7 @@ pub struct QuantParams {
 
 impl QuantParams {
     /// Create symmetric per-tensor quantization params
-    #[must_use] 
+    #[must_use]
     pub fn symmetric_per_tensor(scale: f32) -> Self {
         Self {
             scales: vec![scale],
@@ -151,7 +151,7 @@ impl QuantParams {
     }
 
     /// Create symmetric per-token quantization params
-    #[must_use] 
+    #[must_use]
     pub fn symmetric_per_token(scales: Vec<f32>) -> Self {
         let zero_points = vec![0; scales.len()];
         Self {
@@ -595,7 +595,7 @@ impl SageAttention {
     }
 
     /// Get configuration
-    #[must_use] 
+    #[must_use]
     pub fn config(&self) -> &SageAttentionConfig {
         &self.config
     }
@@ -619,7 +619,7 @@ pub struct SageAttentionBuilder {
 
 impl SageAttentionBuilder {
     /// Create a new builder with default config
-    #[must_use] 
+    #[must_use]
     pub fn new() -> Self {
         Self {
             config: SageAttentionConfig::default(),
@@ -627,35 +627,35 @@ impl SageAttentionBuilder {
     }
 
     /// Set number of heads
-    #[must_use] 
+    #[must_use]
     pub fn num_heads(mut self, num_heads: usize) -> Self {
         self.config.num_heads = num_heads;
         self
     }
 
     /// Set head dimension
-    #[must_use] 
+    #[must_use]
     pub fn head_dim(mut self, head_dim: usize) -> Self {
         self.config.head_dim = head_dim;
         self
     }
 
     /// Set number of KV heads for GQA
-    #[must_use] 
+    #[must_use]
     pub fn num_kv_heads(mut self, num_kv_heads: usize) -> Self {
         self.config.num_kv_heads = Some(num_kv_heads);
         self
     }
 
     /// Enable causal masking
-    #[must_use] 
+    #[must_use]
     pub fn causal(mut self, causal: bool) -> Self {
         self.config.causal = causal;
         self
     }
 
     /// Enable smooth quantization
-    #[must_use] 
+    #[must_use]
     pub fn smooth_quant(mut self, alpha: f32) -> Self {
         self.config.use_smooth_quant = true;
         self.config.smooth_alpha = alpha;
@@ -663,14 +663,14 @@ impl SageAttentionBuilder {
     }
 
     /// Set quantization bits
-    #[must_use] 
+    #[must_use]
     pub fn quant_bits(mut self, bits: u8) -> Self {
         self.config.quant_bits = bits;
         self
     }
 
     /// Use per-token quantization
-    #[must_use] 
+    #[must_use]
     pub fn per_token(mut self, per_token: bool) -> Self {
         self.config.per_token_quant = per_token;
         self