refactor(ml): extract universe, backtesting, asset-selection into sub-crates

- ml-universe (1.7K lines): correlation, liquidity, momentum, volatility
  modules. Only depends on ml-core (MLError, PRECISION_FACTOR).
  6 tests passing.

- ml-backtesting (1.1K lines): action_loader, barrier_backtest, report
  modules. Only depends on ml-core (MLError). Discovered and included
  previously undeclared report.rs module. 10 tests passing.

- ml-asset-selection (1.2K lines): scorer, selector modules with
  AssetClass, AssetUniverse, PredictabilityScorer, ActiveSetSelector.
  Only depends on ml-core (MLError). 33 tests passing.

All three replaced with thin facade re-exports in ml — existing
`use ml::universe::*` / `use ml::backtesting::*` /
`use ml::asset_selection::*` paths continue to work.

Total: 15 sub-crates extracted from ml monolith.
Workspace: 0 errors, ml tests 902 passed + 49 in sub-crates.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

crates/ml/Cargo.toml

@@ -81,6 +81,9 @@ ml-checkpoint.workspace = true
 ml-data-validation.workspace = true
 ml-validation.workspace = true
 ml-risk.workspace = true
+ml-backtesting.workspace = true
+ml-asset-selection.workspace = true
+ml-universe.workspace = true
 config.workspace = true
 common = { workspace = true, features = ["questdb"] }
 risk = { path = "../risk" }