feat(ml): implement TFT layer normalization in quantized GRN

Replace the no-op apply_layer_norm() that returned input unchanged with
real layer normalization using layer_norm_with_fallback. Initialize
LayerNormParams with proper weight (ones) and bias (zeros) tensors in
from_grn constructor instead of None placeholders.

ml/src/tft/quantized_grn.rs

@@ -92,10 +92,15 @@ impl QuantizedGatedResidualNetwork {
         };
 
         // Store layer norm parameters (keep in F32)
+        // Initialize weight=ones and bias=zeros for proper layer normalization
+        let ln_weight = Tensor::ones(grn.output_dim, candle_core::DType::F32, &device)
+            .map_err(|e| MLError::ModelError(format!("Failed to create ln_weight: {}", e)))?;
+        let ln_bias = Tensor::zeros(grn.output_dim, candle_core::DType::F32, &device)
+            .map_err(|e| MLError::ModelError(format!("Failed to create ln_bias: {}", e)))?;
         let layer_norm = Some(LayerNormParams {
             normalized_shape: vec![grn.output_dim],
-            weight: None, // Would extract from grn.layer_norm in production
-            bias: None,
+            weight: Some(ln_weight),
+            bias: Some(ln_bias),
             eps: 1e-5,
         });
 
@@ -221,12 +226,17 @@ impl QuantizedGatedResidualNetwork {
 
     /// Apply layer normalization (kept in F32)
     fn apply_layer_norm(&self, x: &Tensor) -> Result<Tensor, MLError> {
-        // Simplified layer norm for TDD
-        // In production, would use actual layer_norm with weights/bias
-
-        // For now, return input (layer norm would normalize here)
-        // TODO: Implement proper layer normalization
-        Ok(x.clone())
+        if let Some(ln) = &self.layer_norm {
+            crate::cuda_compat::layer_norm_with_fallback(
+                x,
+                &ln.normalized_shape,
+                ln.weight.as_ref(),
+                ln.bias.as_ref(),
+                ln.eps,
+            )
+        } else {
+            Ok(x.clone())
+        }
     }
 
     /// Get quantization type