perf: year-shard index.json to cut initial load from MBs to ~1 year

merge_all/_merged/index.json is now a shard manifest; activities are
split into index-{year}.json files. The feed loads only the most-recent
year on first paint (~200 activities instead of all of them). Older
years are fetched lazily when the user clicks "Load older activities".

Also strips best_efforts / best_climb_m / source from shard files —
these fields are aggregation inputs only, never read by the feed UI.

bincio/render/merge.py

@@ -155,8 +155,7 @@ def merge_one(data_dir: Path, activity_id: str) -> None:
     activities.sort(key=lambda a: a.get("started_at", ""), reverse=True)
     activities.sort(key=lambda a: 0 if a.get("custom", {}).get("highlight") else 1)
 
-    index["activities"] = activities
-    (merged_dir / "index.json").write_text(json.dumps(index, indent=2, ensure_ascii=False))
+    _write_year_shards(merged_dir, activities, index)
 
 
 def merge_all(data_dir: Path) -> int:
@@ -267,11 +266,57 @@ def merge_all(data_dir: Path) -> int:
         activities.sort(key=lambda a: a.get("started_at", ""), reverse=True)
         activities.sort(key=lambda a: 0 if a.get("custom", {}).get("highlight") else 1)
 
-        index["activities"] = activities
-        (merged_dir / "index.json").write_text(
-            json.dumps(index, indent=2, ensure_ascii=False)
-        )
-    elif (merged_dir / "index.json").exists():
-        (merged_dir / "index.json").unlink()
+        _write_year_shards(merged_dir, activities, index)
+    else:
+        # Remove any stale year shard files if the source index disappeared
+        for f in merged_dir.glob("index-*.json"):
+            f.unlink()
+        if (merged_dir / "index.json").exists():
+            (merged_dir / "index.json").unlink()
 
     return len(sidecars)
+
+
+# Fields only needed for athlete.json aggregation at extract time — they add
+# bulk to every summary entry but are never read by the feed UI.
+_FEED_STRIP = {"best_efforts", "best_climb_m", "source"}
+
+
+def _write_year_shards(merged_dir: Path, activities: list[dict], index_meta: dict) -> None:
+    """Split activities by year and write index-{year}.json shards.
+
+    Replaces merged_dir/index.json with a shard manifest so the feed can
+    load only the most-recent year on first paint and fetch older years lazily.
+    """
+    from collections import defaultdict
+
+    # Remove stale year shard files from previous runs
+    for f in merged_dir.glob("index-*.json"):
+        f.unlink()
+
+    by_year: dict[str, list[dict]] = defaultdict(list)
+    for a in activities:
+        year = (a.get("started_at") or "")[:4] or "unknown"
+        # Strip aggregation-only fields to keep shard files small
+        slim = {k: v for k, v in a.items() if k not in _FEED_STRIP}
+        by_year[year].append(slim)
+
+    years = sorted(by_year.keys(), reverse=True)  # newest first
+    shards = []
+    for year in years:
+        shard_doc = {
+            **{k: v for k, v in index_meta.items() if k not in ("activities", "shards")},
+            "shards": [],
+            "activities": by_year[year],
+        }
+        fname = f"index-{year}.json"
+        (merged_dir / fname).write_text(json.dumps(shard_doc, indent=2, ensure_ascii=False))
+        shards.append({"url": fname, "year": int(year) if year.isdigit() else 0,
+                        "count": len(by_year[year])})
+
+    root_doc = {
+        **{k: v for k, v in index_meta.items() if k not in ("activities", "shards")},
+        "shards": shards,
+        "activities": [],
+    }
+    (merged_dir / "index.json").write_text(json.dumps(root_doc, indent=2, ensure_ascii=False))