fix: make service pool selection and routing reliable

[ctkm-aelf]

Summary

• Group the searchable connection selector by original service, with individual accounts and keys underneath. Match renamed labels to Services and search before pagination. Preserve loaded pages, scroll and keyboard focus; finish pending pagination before refreshing compatibility and prevent stale selections.
• Make routing configurable and understandable: show weighted turns and enabled shares, reorder connections globally or within priority tiers, retain drafts across mode changes, and validate the fields used by the selected mode. Explain cycle positions and counter resets.
• Recover from concurrent edits with refreshed pool data and Reload latest. Align Unicode limits between form and API. Preserve unavailable-credential handling, the direct proxy error contract, owner/service scopes and read-only inspection.

Test Plan

• Full frontend suite: 4,246 tests in 429 files; focused pool regressions: 53 passed.
• Production build and lint: no lint errors or new warnings in changed files.
• Five permanent Playwright tests against the production build: grouping, Routing dialogs, short viewports, delayed pagination/selection, retained scroll, keyboard focus, and cached-search compatibility.
• Fresh MongoDB 8 inspection suite: 11 passed, including renamed key labels, platform bindings with retained keys, foreign references, literal name search, pagination, selected/health views and no credential materialization or last-used writes.
• Rust Unicode regression for ASCII, CJK and astral name/description/model boundaries; CLI pool tests: 31 passed.
• Rust formatting, architecture boundary check and diff whitespace check.
• GPT-6-astra xhigh independent re-review: all findings resolved; 16 controlled race/error scenarios and eight cache scenarios passed.
• Claude Opus 5.5 xhigh final re-review: all findings and final presentation observations resolved; nine independent browser regression scenarios and the five permanent browser tests passed. Final loading/failure-state changes passed the 53 focused tests and all five production browser tests.
• Broader service-pool runtime suite: 120 passed on MongoDB 8, covering fallback, streaming/replay limits, cooldowns, ACLs, billing, AI routing and CRUD.
• Final commit 6fc7448f: CI Pipeline, backend/frontend coverage and all CodeQL jobs passed. Backend CI: 7,383 tests passed; frontend CI: 4,246 tests passed. Billing smoke, image-input checks, Clippy, formatting, all KMS feature builds and wizard bundle freshness passed.

CI: https://github.com/ChronoAIProject/NyxID/actions/runs/37222335316

Checklist

• Repository architecture and conventional commit requirements followed.
• API and service-pool documentation updated.
• Metadata reads preserve owner and caller scope.
• Rebased onto main at aed83509 without conflicts. Final CI passed on this base, including the two unrelated catalog/legal-page updates that landed after the full local suite.

Follow-up to #1680 and #1732. Main requires one approving GitHub review before normal squash merge.

[github-actions]

📊 Code coverage

Component	Lines	Threshold	Status	Δ vs base
Backend (nyxid)	87.07%	73%	✅	— 0.00
Frontend (vitest)	72.21%	15%	✅	🔺 +0.11

Gate: line coverage must stay at or above the threshold. Ratchet plan (W21): Backend → 55%, CLI → 50%, Frontend → 30% by quarter end.