[defer] feat(proxy): steer local core selector off set-aside members

[maxmad64bis]

⚠️ base-red inherited: #14547

Summary

Control calls to the local core selector stay invisible when they fail: a blocked target, an unusable secret, a refused switch or a network error only reach the server log, so the operator never sees the steering is broken. Failed control calls now surface as a subscription warning through the existing warning path (status = "ok" + error column carrying SELECTOR_SWITCH_FAILED), persisted in dedicated columns so the next sync does not erase it, and cleared by the next success. The switch slot is reserved before the first await, so two simultaneous triggers on one selector collapse to a single PUT.

Related Issues

No linked issue — failed control calls surfacing as subscription warnings answers the maintainer's #14702 review thread (2026-09-24).

Validation

• Change type: other
• Focused tests and category gates from the golden path
• npm run lint
• Reconciled with the current active release base; focused checks rerun afterward
• Production-code changes include a new or updated automated test in this PR

Tests Added Or Updated

• tests/unit/proxySubscription.selectorWarning.test.ts (new, 3 tests): refused switch → warning persists after sync; next success → cleared; 2 simultaneous triggers → 1 single PUT; 3/3 green (RED-proven both ways).
• Neighbor selector suites re-run green: trigger + functional + guard + schema 27/27.

Coverage Notes

• New warning/merge/slot branches covered by the new suite; no coverage movement beyond the new branches.
• New en.json keys (14, incl. the error code) translated across all 66 locales (check-new-key-coverage: 0 proxySubscription gaps).

Reviewer Notes

• Failed control calls now surface as a subscription warning through the existing warning path (maintainer feat(providers): Apply a local-core entry's set-aside without restarting the core #14702 thread).
• Migration 193 (192 taken by feat(sse): record per-request added wait with cause in call logs #14756); SELECTOR_CONTROL_ALLOWLIST documented in .env.example + ENVIRONMENT.md (check-env-doc-sync green).
• Stacked child fix(proxies): steer selector to members without recent refusals #14804 (fix/selector-member-refusal-memory) sits on the old head — rebase it --onto this head after merge.

Maintainer rework (merge-batch 2026-09-24)

• Merged release/v3.8.51 into the branch (real merge, author commits preserved).
• Dropped the migration-count edits to AGENTS.md, llm.txt, the 66 docs/i18n/*/llm.txt mirrors and README.md: the count was already stale and those files are agent-instruction surfaces that need separate operator approval. They now match the release tip.
• Migration stays 193_proxy_subscription_selector_control.sql (190–192 are reserved for other PRs in the same batch). Note: fix(proxies): steer selector to members without recent refusals #14804 is stacked on this branch and creates the same 193 file.
• Fixed settings.proxySubscription.selectorHelp in all 67 locales: the bare selector=<name> was parsed by next-intl as an unclosed rich-text tag (UNCLOSED_TAG), so the help line fell back to showing the raw key. It is now ICU-escaped ('<name>') and renders literally. Regression test: tests/unit/proxySubscription.selectorHelpIcu.test.ts (red before, green after).

[diegosouzapw]

Migration number coordination: several open PRs claim the same slot, and the release tip is already at 189 (187_proxy_logs_attempts, 188_call_logs_reasoning_encrypted, 189_quota_schedules). This PR's migration keeps 193, so no rename is needed. Slots go by PR age, and merges happen in ascending order (190 → 196) so the numbering gate stays contiguous.