[atsyplenkov]

Summary

index() downloads the AlphaEarth index once and keeps it in the cache.

.index_download
  if the index is already in the cache
    return that file
  curl_download into a temporary file
  .index_check_size
  move the file into the cache

download.file() stops after 60 seconds. The index is too large for that limit, so the download stopped before a complete file reached the cache. curl_download() has no total time limit. The file moves into the cache only after the download finishes and the size matches the server Content-Length. A response with no Content-Length is accepted. A short file is not stored in the cache.

Evidence

• Before: download.file() used the 60 second timeout. A missing Content-Length made the size check stop with "missing value where TRUE/FALSE needed".
  After: pkgload::load_all() and testthat::test_file("tests/testthat/test-index.R") reported 7 passes and 0 failures. The new tests mock the server. One accepts a response with no Content-Length. One rejects a 4 byte file when the server reports 12 bytes. air format and jarl check passed on the edited R files.

Merge Danger

Door: two-way

Revert this commit. An index that is already in the cache is reused. The new download runs only when the cache is empty or overwrite = TRUE.

Blast Radius: the first download

People who already have the index see no change. A new download uses curl instead of download.file().