Add OSDF cache-diagnostic appendix to JRA-3Q heat-index notebook

[hrhampapura]

What & why

Follow-up to the JRA-3Q d640000 read failures debugged in #56. The PelicanFS team confirmed the most likely cause is a cache serving corrupt bytes (deterministic zlib: incorrect header check from Casper; the same chunks decode cleanly from the origin and over HTTPS), and asked us for two things:

1. Print which cache Casper is routed to — they operate many caches and have no easy way to find a bad object in one.
2. Use preferred_caches to pin a known-good cache as the interim lever, until client-side checksumming / automatic failover lands in PelicanFS.

This PR adds a diagnostic appendix to notebooks/jja_heatindex.ipynb. The working direct_reads=True open is left untouched, so the notebook still runs end-to-end.

What's added (appendix, run on Casper)

• Cell A — prints the director-chosen cache for d640000, the full candidate list, and the origin; enables the fsspec.pelican logger so cache selection and Marking cache at <url> as bad events print inline.
• Cell B — re-opens JRA-3Q through the cache (no direct_reads), reproduces the failing read, and dumps get_access_data() to name the cache that served/failed each object (handles both the zlib-corrupt case, recorded as success, and the ContentLengthError case, recorded as failure).
• Cell C — guarded preferred_caches template (keeps "+" for director fallback); inert until a healthy cache host is filled in and USE_PREFERRED_CACHES=True.

Notes

• Diagnostic cells need to be run on Casper to capture the bad cache; their outputs are not committed.
• Verified against pelicanfs 1.3.1 source (get_working_cache/get_origin_url return (url, director_response); get_access_data() keeps the last 3 responses per object path).

🤖 Generated with Claude Code