feat(llm): add Gemini adapter and process book-1-chapter-05

Add GeminiAdapter calling Google's Generative Language REST API
(default model: gemini-2.5-flash). Register "gemini" as third
provider in the factory and CLI. Add rate-limit retry with
exponential backoff to the pipeline's _call_llm helper. Increase
default max_tokens from 2000 to 4096.

Process book-1-chapter-05 via Gemini free tier — 1 new entity
extracted (necessaries-conveniencies-and-amusements-of-life),
41 existing entities correctly skipped by dedup. Canonical set
now at 42 unique entities.

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

examples/infospace-with-history/process_chapters.py

@@ -228,21 +228,34 @@ class ChapterProcessor:
 
     # ── LLM Execution Helpers ─────────────────────────────────────────
 
-    def _call_llm(self, prompt: str, stage_label: str) -> Optional[str]:
+    def _call_llm(self, prompt: str, stage_label: str, max_tokens: int = 4096) -> Optional[str]:
         """Call the LLM and return the content string, or ``None`` on failure.
 
+        Retries up to 3 times on rate-limit (429) errors with exponential backoff.
         Does **not** write any files — callers decide where to persist.
         """
         import time as _time
         from markitect.prompts.execution.models import RunConfig
+        from markitect.llm.exceptions import LLMRateLimitError
 
         print(f"        Calling LLM ({stage_label})...")
         t0 = _time.time()
-        try:
-            response = self.llm_adapter.execute_prompt(prompt, RunConfig())
-        except Exception as exc:
-            print(f"        LLM error ({_time.time() - t0:.1f}s): {exc}")
-            return None
+        max_retries = 3
+        for attempt in range(max_retries + 1):
+            try:
+                response = self.llm_adapter.execute_prompt(prompt, RunConfig(max_tokens=max_tokens))
+                break  # success
+            except LLMRateLimitError as exc:
+                if attempt < max_retries:
+                    wait = 15 * (attempt + 1)  # 15, 30, 45 seconds
+                    print(f"        Rate limited, retrying in {wait}s (attempt {attempt + 1}/{max_retries})...")
+                    _time.sleep(wait)
+                else:
+                    print(f"        LLM rate limit after {max_retries} retries ({_time.time() - t0:.1f}s): {exc}")
+                    return None
+            except Exception as exc:
+                print(f"        LLM error ({_time.time() - t0:.1f}s): {exc}")
+                return None
 
         elapsed = _time.time() - t0
         usage = response.usage
@@ -260,9 +273,9 @@ class ChapterProcessor:
 
         return content
 
-    def _execute_llm(self, prompt: str, output_file: Path, stage_label: str) -> Optional[str]:
+    def _execute_llm(self, prompt: str, output_file: Path, stage_label: str, max_tokens: int = 4096) -> Optional[str]:
         """Call the LLM, write the result to *output_file*, and return it."""
-        content = self._call_llm(prompt, stage_label)
+        content = self._call_llm(prompt, stage_label, max_tokens=max_tokens)
         if content:
             output_file.parent.mkdir(parents=True, exist_ok=True)
             output_file.write_text(content)
@@ -812,7 +825,7 @@ def main():
     parser.add_argument(
         "--provider",
         type=str,
-        choices=["openrouter", "claude-code"],
+        choices=["openrouter", "claude-code", "gemini"],
         default=None,
         help="LLM provider for auto-generating outputs (omit for manual mode)",
     )