fix(ml): migrate remaining files — hyperopt, validation, data_loaders, inference
25+ files fixed:
- hyperopt/adapters: duplicate Arc imports, Device→MlDevice
- validation: regime_analysis rewritten for GpuTensor, ppo_adapter NativeDType
- data_loaders: all 3 loaders migrated to GpuTensor::from_host
- tft/training: MlDevice, GpuAdamW, StreamTensor, CPU loss tracking
- training_pipeline: AdamW→GpuAdamW, NativeDevice fixes
- features/multi_timeframe: removed to_candle_tensor
- inference: ModelForward trait replaces candle_nn::Module
- lib.rs: removed cuda_compat re-export
- ppo/mod.rs: fixed re-export
~95 errors remain in: inference.rs, flash_attention, validation/ppo_adapter,
hyperopt adapter method signatures (blocked on trainable adapter trait).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
c89a1dbf29