chore(ml-backtesting): allow(unsafe_code) on sim/mod.rs for cudarc launches

cudarc kernel launches require `unsafe` blocks — the driver API has no
way to type-check kernel args against the cubin signature. The
workspace-wide `-W unsafe-code` lint produces noise on every launch
site (10+ blocks); suppressing at the module level keeps the lint
useful elsewhere in the crate.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

crates/ml-backtesting/src/sim/mod.rs

@@ -6,6 +6,13 @@
 //!
 //! See docs/superpowers/specs/2026-05-18-real-lob-integration-design.md §2 + §5b.
 
+// cudarc kernel launches are inherently `unsafe`: the driver API has no way
+// to type-check kernel args against the .cubin's signature. Per
+// pearl_no_host_branches_in_captured_graph + the surrounding GPU-contract
+// pearls, all launches are pre-compiled and the args are matched by hand in
+// review. The workspace-wide `-W unsafe-code` lint is therefore noise here.
+#![allow(unsafe_code)]
+
 mod batched_config;
 pub use batched_config::{BatchedSimConfig, ResolvedSimVariant, UniformSimParams};