Don't cache BGG's transient "please try again" collection response

[MSGuzy]

Summary

CachedHttpClient.get() in scripts/gamecache/http_client.py caches every successful HTTP response unconditionally - including BGG's async "please try again later" placeholder message that the collection endpoint returns while it prepares a stats=1 export.

The bug

1. BGG's /collection endpoint returns HTTP 202 with:

   <message>Your request for this collection has been accepted and will be processed. Please try again later for access.</message>

   while it builds the export. urlopen doesn't treat 202 as an error, so make_http_request returns this message as a normal response body.
2. CachedHttpClient.get() writes it into the SQLite cache with status_code = 200 and the full expire_after TTL (24h via Downloader).
3. BGGClient._make_request sees the "has been accepted" message and retries via sleep_with_backoff_and_jitter - but every retry hits the same cache key, which now holds the placeholder, so it just keeps replaying "still processing" instead of ever re-checking BGG.
4. Because the retry backoff itself grows unboundedly (10 * 2^tries), this can look like download_and_index.py --cache_bgg has hung - in practice I saw it loop for over an hour, while a direct request to BGG (bypassing the local cache) was returning HTTP 200 with the full collection the whole time.

Fix

Skip the cache write when the response body is BGG's transient message, so every retry actually reaches BGG instead of arguing with its own cache.

Testing

Reproduced the hang locally (collection stuck retrying against a cached placeholder for well over an hour), applied this fix, cleared the cache, and confirmed a fresh run picks up BGG's real response immediately once it's ready, instead of re-caching the placeholder.