feat: validate the selected SOXL research candidate

[Pigbibi]

The completed SOXL manual study could produce development results but could not validate its selected three-asset candidate. Add an explicit soxl_validation operation to the existing workflow, bound to the original completed summary and the published UESP consumer from #490. It compares only 0.55 with 0.65 at the original three costs, retains the development folds, and fixes the independent OOS window to 2025-08-04 through 2026-08-04.

The numeric strategy runtime stays at its original lock. A separate QPK 7363011 control environment applies the existing strict promotion gate and rejects conflicting windows, result identities, costs, or authority declarations. The artifact reports the drawdown/return tradeoff with a pending human quality decision. The operation makes no new AI call and grants no promotion or trading authority; the default readback and existing learning paths retain their behavior.

Validation: 47 focused tests, full Ruff, actionlint, and whitespace checks passed. A fresh Python 3.11 environment imported the pinned gate successfully; synthetic results from the actual QPK 5c916 orchestrator passed all six control-side QPK 7363011 checks. Local broader testing produced 1042 passed, 1 skipped, 280 subtests passed, and 12 failures solely from the missing installed ai_gateway_client package; CI installs the established SDK dependencies and runs the full suite. No real OOS computation, workflow dispatch, broker call, or cloud configuration change was performed.