fix(ml): migrate all test code to GPU types — 125 compile errors fixed
17 files: smoke_tests (gpu_residency, training_stability, performance,
helpers), inference, multi_timeframe, tlob, validation/adapters,
tft/tests, ensemble/adapters/dqn, flash_attention, hyperopt/tft,
mamba/trainable, online_learning, liquid/adapter, transformers/attention
- Tensor→CudaSlice direct ops in smoke tests (no tensor_to_cuda_slice)
- Device→MlDevice throughout
- GpuLinear::forward 4-arg signature (input, store, cublas, stream)
- rand::rng()→thread_rng() (rand 0.8)
- StreamTensor .shape field access (not .dims() method)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
7e49d3c08e