feat: series-level voice-drift finding for uniformly off-register corpus (#2248)

[atomantic]

Closes #2248

Summary

Follow-up to the #2179 voice-fingerprint baseline hook (PR #2247). Closes the residual boundary in the style.voice-drift check's exemplar/blended baseline mode: when every drafted issue shares the same value on a metric (drafted σ≈0) but that shared value sits far from the chosen-voice center, the per-issue z-score model emits nothing — there's no per-issue outlier, the whole corpus is uniformly off-register. The canonical case is a dialogue-free series (dialogueRatio = 0 on every issue) measured against a dialogue-heavy chosen voice.

computeVoiceDrift now emits a distinct series-level finding for exactly this case: a metric with σ≈0 whose |center − mean| exceeds a scale-free relative distance threshold (|center − mean| / max(|center|, |mean|) ≥ 0.5 by default). Scale-free because the metric vector mixes wildly different scales (dialogueRatio 0–100, sentence-length CV ~0.4, em-dash rate per-1k) — a single absolute gap threshold would be meaningless.

What shipped

• server/lib/editorial/voiceFingerprint.js — computeVoiceDrift returns a new seriesFindings[] (sorted by distance desc), populated only under an active exemplar/blended baseline at the σ≈0 skip site (drafted mode never trips it — center === mean). New describeSeriesDrift() renders the finding sentence ("The whole series sits at 0% on dialogue ratio vs the style guide's chosen voice of 42.5% — the corpus is uniformly below the chosen voice here").
• server/lib/editorial/checks/proseStyle.js — style.voice-drift run() maps seriesFindings into findings with issueNumber: null (a corpus-wide observation, not an outlier), a Series-wide location, escalated one severity rank above the per-issue floor. They share the maxFindings cap, listed first so a series-wide finding is never starved by a flood of per-issue ones.
• server/services/pipeline/voiceFingerprint.js — surfaces seriesFindings on the matrix-view payload.
• client/src/pages/PipelineVoiceFingerprint.jsx — renders a banner above the matrix for series-wide mismatches (they have no single cell to highlight since the column is uniform).
• Docs: server/lib/editorial/README.md row + .changelog/NEXT.md entry under Creative Writing Quality Engine.

Test plan

• server/lib/editorial/voiceFingerprint.test.js — new computeVoiceDrift series-level findings (#2248) block (drafted-mode never emits; flags the uniformly-off-register metric; distance-desc sort; custom threshold; σ>0 metrics stay in the per-issue path; gatedOff → empty; describeSeriesDrift text) + a registry-wiring test asserting issueNumber: null / Series-wide / escalated severity.
• client/src/pages/PipelineVoiceFingerprint.test.jsx — banner renders for seriesFindings, absent when empty.
• ✅ server editorial suite (912 passed) + client fingerprint page suite (8 passed) green.
• Pre-existing DB-backed pipeline suites (series.test.js, issues.test.js, etc.) fail locally on both this branch and clean main because the portos_test DB isn't provisioned in this environment — unrelated to this change.