feat(signals): deterministic slop-assessment core

[JSONbored]

Parent roadmap: #525
Parent phase: #526

Background

A contributor's AI harness can write code but cannot judge it against Gittensor's contribution rules. We need a single deterministic scorer for "AI slop vs real contribution" that both the miner MCP tool (#526) and the maintainer CI consume, so the verdict is consistent on both sides.

Goal

Add buildSlopAssessment(input) -> { slopRisk: 0..100, band: clean|low|elevated|high, findings: SignalFinding[] } plus a SLOP_RUBRIC_MARKDOWN constant, in a new src/signals/slop.ts.

Current Behavior

No slop logic exists (grep slop src is empty). Readiness/preflight signals exist but there is no slop score.

Desired Behavior

A pure, deterministic, metadata-only scorer. Signals, each mapped to an existing builder:

• missing test evidence → reuse src/signals/test-evidence.ts
• trivial / whitespace-only churn → reuse buildLocalScoreInput token split
• generated/vendored edits (dist/, *.min.*, *.lock, vendor/, *.generated.*) → new path matcher
• dependency-bump-only; docs-only padding → new path matchers
• no linked issue without rationale → reuse preflight missing_linked_issue / hasClearNoIssueRationale
• duplicate-cluster membership → reuse buildCollisionReport (src/signals/engine.ts)
• low-quality commit messages → reuse STOPWORDS

Reuse the existing SignalFinding shape so output flows through current renderers/redactors unchanged.

Implementation Requirements

• New src/signals/slop.ts; weights in one tunable SLOP_WEIGHTS constant; no clock/random in the scoring math.
• Reuse isTestFile/isCodeFile from src/signals/local-branch.ts; do not re-implement file classification.

Public/Private Output Boundaries

• slopRisk is a quality number (like readiness.total), NOT a trust/reward score.
• Every public-facing reason string must pass the shared public-safe filter (see roadmap(pillar 4): foundations & learning loop #529 redaction module).

Acceptance Criteria

• Identical metadata input → identical slopRisk (deterministic). Bands behave at documented thresholds.
• Findings render through existing public surfaces without new private-term leakage.

Testing Requirements

• npm run test:ci, 97%+ coverage. Golden-fixture + property tests mirroring buildPublicReadinessScore / buildScorePreview. Forbidden-term assertions on all reason strings.

Sub-issues

• feat(signals): slop core — scaffold buildSlopAssessment #558 scaffold buildSlopAssessment
• feat(signals): slop signal — missing test evidence #559 missing test evidence
• feat(signals): slop signal — trivial/whitespace churn #560 trivial/whitespace churn
• feat(signals): slop signal — generated/vendored/dependency/docs matchers #561 generated/vendored/dependency/docs matchers
• feat(signals): slop signal — no linked issue without rationale #562 no linked issue without rationale
• feat(signals): slop signal — duplicate-cluster membership #563 duplicate-cluster membership
• feat(signals): slop signal — low-quality commit messages #564 low-quality commit messages
• feat(signals): slop weight tuning + golden-fixture tests #565 weight tuning + golden-fixture tests