fix(cuda+csm): P0 rtx-backend-cuda compile fix + P2 rtx-csm clippy cleanup (#9)

Co-authored-by: Omar Sobh <[email protected]>
Co-committed-by: Omar Sobh <[email protected]>

crates/models/rtx-csm/src/training.rs

@@ -385,7 +385,7 @@ impl<'a> Trainer<'a> {
                     self.generator.model.inner.refresh_lora(self.vm)?;
 
                     step += 1;
-                    if step % 10 == 0 || step == total_steps - 1 {
+                    if step.is_multiple_of(10) || step == total_steps - 1 {
                         tracing::info!(
                             "  step {step:>4}/{total_steps} epoch {epoch} ex {idx} frame {frame_idx} \
                              prompt_len={prompt_len} lr={lr:.2e} loss={loss_val:.4}"
@@ -407,9 +407,11 @@ impl<'a> Trainer<'a> {
 ///   - any other string matches `ex.stage == Some(label)` exactly
 ///
 /// Per `personal_voice_training_guide.md` §4 the canonical 3-stage recipe is:
+///
 ///   - audiobook: 3 epochs, peak_lr 1e-4
 ///   - podcast:   1 epoch,  peak_lr 3e-5
 ///   - va:        1 epoch,  peak_lr 1e-5
+///
 /// each over a growing pool. The `CurriculumTrainer` enforces the pool growth
 /// implicitly by virtue of how you label your manifest (a podcast example
 /// labeled `stage: "podcast"` is only matched by the podcast stage; if you