fix(wiki): LLM error handling — graceful degradation on Ollama unavailable

Wrap llm.generate() in compile_concept with a broad try/except;
log a warning and return an empty ConceptPage (no crash) when the
LLM endpoint is unreachable. Add logging module import.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

codex/wiki.py

@@ -11,12 +11,14 @@ Compile-state (hash tracking for incremental re-runs) is persisted to
 Graceful degradation:
 - Missing ``formulas`` table (F-09 not present) → no formula embedding, no crash.
 - Missing ``verify_citations`` (F-10 not present) → local substring grounding check.
+- LLM unavailable (httpx.ConnectError or any exception) → empty ConceptPage, no crash.
 """
 
 from __future__ import annotations
 
 import hashlib
 import json
+import logging
 import re
 import textwrap
 from dataclasses import dataclass, field
@@ -28,6 +30,8 @@ import yaml
 
 from codex.config import get_settings
 
+logger = logging.getLogger(__name__)
+
 # ---------------------------------------------------------------------------
 # Dataclasses
 # ---------------------------------------------------------------------------
@@ -442,7 +446,11 @@ def compile_concept(
 
     h = _chunk_hash(chunks)
     prompt = _build_synthesis_prompt(concept, chunks)
+    try:
         raw_output = llm.generate(prompt, model=settings.wiki_llm_model)
+    except Exception as exc:  # noqa: BLE001
+        logger.warning("LLM unavailable (%s): skipping concept %s", exc, concept.slug)
+        return ConceptPage(concept=concept, markdown="", claims=[], chunk_hash=h)
 
     claims = _parse_claims(raw_output)
     claims = _run_grounding_guard(claims, chunks)