fix: the ZIP restore that failed main, the Zenodo survey's question size, and lettered-star spellings

[alowpoly]

Three fixes, one of them for a failure this session put on main.

• ZIP restore (failing on main since feat(exoplanets): WASP-94 A b opens on its published climate model, read from a NetCDF-4 release #1357). The streamed zip-member step pinned an empty file when unzip finished before the write began to read it: Node discards a finished child's unread output. Its output is now piped on at once. Reproduced locally: 0 bytes read late, all 22 bytes when piped at once.
• Scaffold test (failing on main since feat(stars): add the central stars of the Ring, Owl, Helix and Southern Ring nebulae #1352). The test still expected the 40,000 K refusal; the code says 250,000 K.
• Zenodo survey. The generator's survey asked about 40 planets in one question, which Zenodo now answers with HTTP 500, so every draft run reported "Zenodo could not be asked". Questions are now sized by the spellings they ask for (32 at most). Planets of lettered stars gain their spellings: Zenodo writes "WASP-94A b", which went unmatched.

Proves the change: measured on 5 October 2026 over 110 questions: every one of up to 33 spellings was answered, every one of 44 or more refused. The fixed survey, live, over all 896 exoplanets of the tree: 70 questions, 71 requests, 201 s, no failure, 26 planets with records, WASP-94 A b and TOI-1259 A b among them.

Limits: Zenodo documents no limit, so 32 is a measured size.

Clean: operations-acquisition, new-object-cli, simulation-dataset, simulations, leads and published-map-dataset tests; ESLint on changed files.

PR rules.