fix(ci): hydrate script crashes on snapshot entries with null byte sizes

The Raincloud conformance workflow failed on its first real run: the
size-cap computation used entry.get('parquet_bytes', 0), but .get returns
the default only for ABSENT keys — a snapshot entry with an explicit
"parquet_bytes": null (a not-yet-hashed slug: jsonbench-bluesky-100m,
wikipedia-structured-contents, code-contests, bi-commongovernment) returns
None, and None + int raised TypeError, aborting the whole hydrate step.
Coerce both fields with 'or 0' so missing and null both cap as 0.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: dfa1

scripts/hydrate-raincloud-corpus.sh

@@ -77,8 +77,11 @@ import raincloud
 
 max_bytes = int(os.environ["MAX_MB"]) * 1024 * 1024
 snapshot = json.loads(resources.files("raincloud").joinpath("_data/snapshot.json").read_text())
+# `.get(key, 0)` returns 0 only when the key is absent; a snapshot entry can carry
+# an explicit "parquet_bytes": null (not-yet-hashed slug), which .get returns as None.
+# `or 0` coerces both the missing and the null case to 0 so the size cap still applies.
 sizes = {
-    slug: entry.get("parquet_bytes", 0) + entry.get("vortex_bytes", 0)
+    slug: (entry.get("parquet_bytes") or 0) + (entry.get("vortex_bytes") or 0)
     for slug, entry in snapshot["slugs"].items()
 }