feat(ml-backtesting): batched-cell sweep schema + 140-variant runner (P6)

Sweep YAML now supports the batched flow per spec §3.3 + Task 6:
- SweepBase.sim_variants: Vec<SimVariant> — list of (cost, latency,
  threshold, ...) variants. When non-empty, each cell runs ONE harness
  at n_parallel=variants.len() with BatchedSimConfig::from_grid instead
  of the legacy one-harness-per-cell fan-out.
- SweepBase.data_template: Option<String> — when set with `{window}`
  placeholder, each cell's `window` field interpolates the per-cell
  data path. Replaces single scalar `data` for the windowed flow.
- SweepCell.window: Option<String> — window identifier (e.g., "2025-Q2").
- SimVariant: threshold + cost_per_lot_per_side required (the spec's
  primary axes); other fields optional overrides on top of SweepBase
  scalars.

New runner pieces:
- BatchedSimConfig::from_grid(&[ResolvedSimVariant]) in
  crates/ml-backtesting/src/sim/batched_config.rs.
- ResolvedSimVariant — per-variant fully-resolved sim params.
- resolve_sim_variants(&SweepBase) in main.rs — layers per-variant
  overrides over base scalars.
- run_batched_cell() in main.rs — builds the harness with
  sim_config_override + variant_names plumbed through. Writes per-
  backtest artifacts to sim_<variant_name>/ subdirs (spec §3.3).

Harness side:
- BacktestHarnessConfig gains variant_names + sim_config_override
  Option fields. When sim_config_override is Some, harness uses that
  directly instead of building from_uniform off scalar cfg. When
  variant_names is Some, write_artifacts uses sim_<name>/ instead of
  cell_NNNN/ subdirs. Both None preserve legacy single-cell behaviour
  (smoke, fixtures unchanged).

YAML configs:
- config/ml/sweep_threshold_tuning.yaml: 1 cell (W0) × 8 sim_variants
  (p60-p95 in 5pt steps) with cost=0.125 (1-tick anchor). Threshold
  pre-registration pass.
- config/ml/sweep_deployability.yaml: 4 cells (W1-W4) × 140 variants
  each (7 costs × 4 latencies × 5 thresholds). Generated by
  scripts/generate_sweep_variants.py — placeholder threshold values
  (p60-p95) until threshold-tuning publishes calibrated absolutes to
  config/ml/v2_prod_thresholds.json.

Deferred to P7 (operational glue):
- argo-lob-sweep.sh adaptation for the batched flow (cells = windows,
  not sim-variants; one Argo task per window invokes `fxt-backtest sweep`
  end-to-end inside the pod rather than `fxt-backtest run`).

Regression: all 7 existing CUDA tests pass through the new harness
construction path (sim_config_override = None → from_uniform fallback).

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

scripts/generate_sweep_variants.py

@@ -0,0 +1,75 @@
+#!/usr/bin/env python3
+"""P6 helper: emit the 140 sim_variants for the deployability sweep.
+
+Cross-product:  7 costs × 4 latencies × 5 thresholds = 140 variants.
+
+Threshold values are placeholders — the actual values come from
+config/ml/v2_prod_thresholds.json after the threshold-tuning step.
+Re-run this script post-tuning to refresh sweep_deployability.yaml with
+calibrated thresholds.
+
+Usage:
+  scripts/generate_sweep_variants.py > config/ml/sweep_deployability.yaml
+"""
+from __future__ import annotations
+import json, sys
+from pathlib import Path
+
+# 7 costs (price-units / lot / side). 0.125 = 1 tick (ES), 0.0625 = 0.5 tick (best-case),
+# stepping through realistic retail (0.25, 0.375 = 2-3 ticks), and worst-case (1.0 = 8 ticks).
+COSTS = [0.0625, 0.125, 0.1875, 0.25, 0.375, 0.5, 1.0]
+
+# 4 latencies (ns). 100M = 100ms (best), 200M = realistic anchor, 300M = degraded,
+# 400M = stress anchor.
+LATENCIES_NS = [100_000_000, 200_000_000, 300_000_000, 400_000_000]
+
+# 5 thresholds — REPLACE THESE WITH THE OUTPUT OF THE THRESHOLD-TUNING PASS.
+# Read from config/ml/v2_prod_thresholds.json if it exists; otherwise use
+# the canonical p60-p95 placeholder (5 steps spanning the range).
+THRESH_JSON = Path(__file__).resolve().parent.parent / "config" / "ml" / "v2_prod_thresholds.json"
+if THRESH_JSON.exists():
+    blob = json.loads(THRESH_JSON.read_text())
+    THRESHOLDS = blob["thresholds_absolute"]   # e.g. {"p60": 0.42, "p70": 0.55, ...}
+    THRESHOLD_NAMES = list(THRESHOLDS.keys())
+    THRESHOLD_VALUES = list(THRESHOLDS.values())
+    assert len(THRESHOLD_VALUES) == 5, f"expected 5 thresholds; got {len(THRESHOLD_VALUES)}"
+else:
+    print("# WARNING: v2_prod_thresholds.json missing — using p60-p95 placeholder thresholds",
+          file=sys.stderr)
+    THRESHOLD_NAMES  = ["p60",  "p70",  "p80",  "p90",  "p95"]
+    THRESHOLD_VALUES = [0.40,   0.55,   0.70,   0.85,   0.95]
+
+assert len(COSTS) == 7
+assert len(LATENCIES_NS) == 4
+
+print("# P6 deployability sweep — 4 windows × 140 sim_variants (7 cost × 4 latency × 5 threshold).")
+print("# Generated by scripts/generate_sweep_variants.py — do NOT hand-edit.")
+print("# Re-run after threshold-tuning to refresh threshold absolute values.")
+print()
+print("base:")
+print("  data_template: /mnt/training-data/futures-baseline-mbp10/ES.FUT/ES.FUT_{window}.dbn.zst")
+print("  predecoded_dir: /feature-cache/predecoded")
+print("  n_parallel: 1                        # batched flow overrides with variants.len()=140")
+print("  decision_stride: 4")
+print("  latency_ns: 200000000                # default (overridden per variant)")
+print("  target_annual_vol_units: 50.0")
+print("  annualisation_factor: 825.0")
+print("  max_lots: 5")
+print("  max_events: 0")
+print("  seed: 12648430")
+print("  # __SHA__ replaced by submitter with post-trunk-grows training short-SHA.")
+print("  checkpoint: /feature-cache/alpha-perception-runs/__SHA__/trunk_best_h6000.bin")
+print("  sim_variants:")
+
+for ci, cost in enumerate(COSTS):
+    for li, latency in enumerate(LATENCIES_NS):
+        for ti, (tname, tval) in enumerate(zip(THRESHOLD_NAMES, THRESHOLD_VALUES)):
+            name = f"c{ci}_l{li}_{tname}"
+            print(f"    - {{ name: {name}, threshold: {tval}, cost_per_lot_per_side: {cost}, latency_ns: {latency} }}")
+
+print()
+print("cells:")
+print("  - { name: W1, window: 2025-Q2 }")
+print("  - { name: W2, window: 2025-Q3 }")
+print("  - { name: W3, window: 2025-Q4 }")
+print("  - { name: W4, window: 2026-Q1 }")