feat(infospace,llm): stabilize free-tier eval workflow

Five improvements that eliminate most of the agent-in-the-loop friction
observed while closing out the 988-entity WoN evaluation (C.1):

1. Gemini adapter now retries on 429 + 5xx with exponential backoff
   (same pattern already used by OpenRouter/OpenAI). Removes the need
   for shell-level retry wrappers when hitting free-tier rate limits.

2. evaluate CLI prints the underlying error ("ERROR — HTTP 503 …")
   instead of a bare "ERROR", so agents don't have to drop into Python
   to diagnose transient failures.

3. --entity/--chapter now respect existing evaluation files by default
   (previously only the full-collection pass did). New --force flag
   opts into re-evaluation. Stops silently burning free-tier quota on
   re-runs of the same slug.

4. --entity accepts hyphenated slugs (matching entity filenames) and
   normalizes them to the underscore form used on disk. On a miss the
   CLI suggests near matches instead of a bare "not found".

5. eval-summary --update-metrics is no longer destructive:
   read_metrics_file/write_metrics_file preserve structured values
   (type_distribution) and don't flatten ints to floats. Fixes a
   silent data loss observed on every run.

Bonus: the evaluator field in written evaluation frontmatter now
falls back from run_config.model_name to the adapter's resolved model
(or the model echoed back in the API response), so rows no longer
show `evaluator: null` when --model is omitted.

Tests: new tests/unit/llm/test_gemini.py covers retry behavior;
tests/unit/infospace/test_history.py gains a round-trip test that
pins the type_distribution / int-preservation invariants.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

markitect/llm/gemini.py

@@ -9,7 +9,11 @@ from markitect.llm.adapter import LLMAdapter
 from markitect.llm.models import RunConfig, LLMResponse
 from markitect.llm.config import resolve_api_key, find_project_root
 from markitect.llm._http import post_json
-from markitect.llm.exceptions import LLMConfigurationError
+from markitect.llm.exceptions import (
+    LLMConfigurationError,
+    LLMAPIError,
+    LLMRateLimitError,
+)
 
 _DEFAULT_MODEL = "gemini-2.5-flash"
 _API_BASE = "https://generativelanguage.googleapis.com/v1beta"
@@ -26,10 +30,12 @@ class GeminiAdapter(LLMAdapter):
         model: Optional[str] = None,
         api_key: Optional[str] = None,
         system_prompt: Optional[str] = None,
+        max_retries: int = 3,
         **_kwargs: Any,
     ):
         self._model = model or _DEFAULT_MODEL
         self._system_prompt = system_prompt
+        self._max_retries = max_retries
 
         root = find_project_root()
         key_file_paths = [root / "apikey-geminifree.txt"] if root else []
@@ -77,7 +83,7 @@ class GeminiAdapter(LLMAdapter):
         url = f"{_API_BASE}/models/{model}:generateContent?key={self._api_key}"
 
         start = time.time()
-        data = post_json(url, payload, timeout=config.timeout_seconds)
+        data = self._post_with_retries(url, payload, timeout=config.timeout_seconds)
         latency = time.time() - start
 
         # Parse Gemini response
@@ -113,3 +119,27 @@ class GeminiAdapter(LLMAdapter):
         if not (0.0 <= config.temperature <= 2.0):
             return False
         return True
+
+    # ── Internals ───────────────────────────────────────────────────
+
+    def _post_with_retries(
+        self,
+        url: str,
+        payload: Dict[str, Any],
+        timeout: int,
+    ) -> Dict[str, Any]:
+        last_exc: Optional[Exception] = None
+        for attempt in range(self._max_retries + 1):
+            try:
+                return post_json(url, payload, timeout=timeout)
+            except LLMRateLimitError as exc:
+                last_exc = exc
+                if attempt < self._max_retries:
+                    time.sleep(2 ** attempt)
+            except LLMAPIError as exc:
+                if exc.status_code in (502, 503, 504) and attempt < self._max_retries:
+                    last_exc = exc
+                    time.sleep(2 ** attempt)
+                else:
+                    raise
+        raise last_exc  # type: ignore[misc]