fix: backtest evaluator action_select kernel arg mismatch (SIGSEGV)

The action_select kernel signature changed from single `float epsilon`
to 4 args (eps_start, eps_end, current_epoch, total_epochs) for GPU-side
cosine schedule. The backtest evaluator wasn't updated, causing a
segfault from misaligned kernel args.

Fix: pass eps_start=0, eps_end=0 (greedy in backtest, no exploration).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/ml/src/cuda_pipeline/gpu_backtest_evaluator.rs

@@ -1031,7 +1031,6 @@ impl GpuBacktestEvaluator {
         let b2 = dqn_cfg.branch_2_size as i32;
         let v_min_f = dqn_cfg.v_min;
         let v_max_f = dqn_cfg.v_max;
-        let epsilon = 0.0_f32;
         let state_row_bytes = n * state_dim * std::mem::size_of::<half::bf16>();
 
         let total_chunks = (self.max_len + DQN_BACKTEST_CHUNK_SIZE - 1) / DQN_BACKTEST_CHUNK_SIZE;
@@ -1132,7 +1131,10 @@ impl GpuBacktestEvaluator {
                     .arg(ch_actions)
                     .arg(ch_rng)
                     .arg(ch_q_gaps)
-                    .arg(&epsilon)
+                    .arg(&0.0_f32)     // eps_start (greedy in backtest)
+                    .arg(&0.0_f32)     // eps_end
+                    .arg(&0_i32)       // current_epoch
+                    .arg(&1_i32)       // total_epochs
                     .arg(&batch_i32)
                     .arg(&b0)
                     .arg(&b1)