Connect eligible SOXL watcher diagnoses to bounded VPS learning

[Pigbibi]

The scheduled strategy watcher now hands an eligible SOXL V3 diagnosis to the existing VPS runner for one bounded learning experiment. It reads only the task-bound frozen P1 input, invokes the original pinned numerical consumer, and records the sanitized nine-result terminal state in the existing research Issue. Trusted terminal records are reused; incomplete or conflicting prior attempts prevent replay.

The dependent job is gated before enqueue by the exact repository, main branch, schedule event, and two dedicated WIF variables. It keeps research Codex-only and grants no order or promotion authority.

Validation:

• Full contributor unittest check in an isolated CI-matched dependency environment: 743 tests, 1 skipped; dependency check and Ruff passed.
• Focused regressions: 85 passed, 16 subtests; workflow actionlint passed.
• One synthetic frozen P1 integration invoked the original pinned numerical CLI once and produced all nine bounded results. This is integration evidence, not a new real-market backtest.

Production follow-up is limited to a separate scheduled-watcher read-only WIF identity and enabling the existing VPS runner. The previous successful manual study is not replayed; no new scheduler, paid API fallback, or trade activation is added.