Resolves #13851

Related: #13874, #13923

Adds the missing post-load LM Studio readiness check: after ensureLmsModelsLoaded() reloads context-configured models, it now verifies observed lms ps --json metadata for loaded context length and configured chat parallel slots before declaring LMS ready. Plain presence-only readiness remains unchanged when no context/parallel gate is configured.

Evidence: L2 (mocked lms ps --json dispatch + focused provider-readiness unit coverage) plus a non-mutating live L3 readiness probe. The live probe confirms the corrected burn-source risk: current LM Studio reports text-embedding-qwen3-embedding-8b loaded at 8192 while Neo's config requires 32768, and the rebased helper flags that model as insufficient. Residual: operator may still run the issue's heavy-session digest / choke-sweep post-merge.

Deltas from ticket

• Uses the existing orchestrator task path through buildLmsPreloadConfig() and ensureLmsModelsLoaded() instead of adding another orchestrator layer.
• Verifies the configured chat --parallel slot count when present, preventing LM Studio's default multi-slot KV-cache split from satisfying readiness.
• Keeps lms ps --json narrowly gated to LMS models with configured contextLengths or parallels, so deployments without context enforcement do not gain a new CLI dependency.
• Confirms the embedding-on-ingestion caveat from #13923 separately from the REM/chat retry-loop fix in #13922.

Contract Ledger

Test Evidence

• node --check ai/services/graph/providerReadinessHelper.mjs
• node --check test/playwright/unit/ai/scripts/runners/runSandman.spec.mjs
• npm run test-unit -- test/playwright/unit/ai/scripts/runners/runSandman.spec.mjs --workers=1 - 47 passed
• node --input-type=module -e '<live lms ps comparator>' - flagged text-embedding-qwen3-embedding-8b observed 8192 required 32768
• npm run agent-preflight -- ai/services/graph/providerReadinessHelper.mjs test/playwright/unit/ai/scripts/runners/runSandman.spec.mjs
• node buildScripts/util/check-block-alignment.mjs ai/services/graph/providerReadinessHelper.mjs test/playwright/unit/ai/scripts/runners/runSandman.spec.mjs
• git diff --check origin/dev..HEAD

Post-Merge Validation

• On a host with LM Studio running, rerun the issue's operator command sequence: lms unload gemma-4-31b-it && lms load gemma-4-31b-it --context-length 131072 --parallel 1, then verify the choke sweep stops failing at the former 4K boundary.
• Verify one real heavy session digests to a non-null tri-vector after the readiness gate is active.

Authored by Euclid (GPT-5, Codex Desktop).