Build a real production runSlopAssessment implementation

[JSONbored]

Problem: attempt-runner.js's injectable deps.runSlopAssessment (packages/gittensory-miner/lib/attempt-runner.js:23) has no production implementation anywhere in this package — the only non-type-signature reference outside the module itself is the test double in test/unit/miner-attempt-runner.test.ts:110 (runSlopAssessment: () => noopSlop). The real self-review scorer (src/signals/slop.ts) is far larger and more interconnected with the main review-stack app, and was never adapted into a portable, injectable form the miner's iterate-loop can call.

Area: AMS / Miner

Proposal: Build a real runSlopAssessment implementation reusing src/signals/slop.ts's scoring logic (or its already-portable packages/gittensory-engine equivalent, if one exists — check before assuming it doesn't) as an injectable dependency conforming to attempt-runner.js's expected shape, so a real attempt gets a genuine self-review pass instead of a no-op.

Deliverables:

• A production runSlopAssessment function, wired as the default dependency for the CLI dispatch in the sibling issue.

Acceptance criteria:

• Real assessment output on a real diff is provably equivalent to (or a documented, deliberate subset of) src/signals/slop.ts's judgment — not a reimplementation that can drift from the live gate's own scoring.
• Existing test-double behavior in miner-attempt-runner.test.ts is unaffected — the test double stays a test double, this issue only adds the real production path.

Resources:

• src/signals/slop.ts (the scorer to reuse).
• packages/gittensory-engine (check first whether this logic is already extracted here for byte-parity, per the project's established "engine = shared deterministic logic" pattern, before building a second copy).

Boundaries:

• Reuse, don't reinvent scoring logic — this is an adapter/wiring task, not a new scoring model.

---

Part of #5130.

[linear-code]

JSO-3681