Refuse a resampling gdalcubes would read as near (#96)

[NewGraphEnvironment]

Summary

• dft_stac_cube(), dft_stac_composite() and dft_stac_fetch() now refuse a resampling that gdalcubes::cube_view() would silently read as "near". Before, resampling = "bilinaer" returned and cached a nearest-neighbour cube with no warning. Fixes resampling typo silently becomes "near" in dft_stac_cube / dft_stac_composite / dft_stac_fetch #96, the same mechanism dft_stac_composite(aggregation = "count") silently returns reflectance, not a clear-observation count #92 fixed for aggregation.
• Allowed set, measured: each value was round-tripped through cube_view()$resampling on gdalcubes 0.7.5, rather than taken from the docs. Twelve come back unchanged: near bilinear cubic cubicspline lanczos average mode max min med q1 q3.
  • mean and median are honoured but renamed, to average and med. They are refused, and the error names the spelling to use.
  • Everything else becomes near, including GDAL's own sum, rms and gauss.
  • Case is ignored, and the value is hashed as given, so no existing cache key moves.
• Where it is checked: before any network call in all three entry points, and again at both cube_view() call sites (stac_cube_assemble(), fetch_extent_to()). fetch_extent_to() also gains the dft_stac_composite(aggregation = "count") silently returns reflectance, not a clear-observation count #92 aggregation_check() it lacked. Both checks share one helper, cube_view_choice_check().
• Release v0.21.0, not a patch: refusing mean/median breaks calls that used to give correct output.

Related Issues

• Fixes resampling typo silently becomes "near" in dft_stac_cube / dft_stac_composite / dft_stac_fetch #96
• Relates to NewGraphEnvironment/sred-2025-2026#16

Test plan

• devtools::test(): FAIL 0 | PASS 1508 | SKIP 16
• Behaviour pin: every allowed value round-trips through cube_view() unchanged, bilinaer comes back near, and the aliases come back as their canonical names
• Mutation table, 11 of 11 killed: dropping any one of the five checks, widening or narrowing the set, lower-casing the return, dropping the alias hint, or naming the wrong fallback each turns a test red
• /code-check: round 1 found one vacuous assertion ("near" also matched the allowed set printed in the headline), now fixed. Rounds 2 and 3 were clean, and a docs fact-check round was clean
• devtools::document() and lintr clean on the touched files

Notes

• BULK scale test not run: this is argument validation before any read, so it changes no raster computation and has no memory or runtime surface.
• Old caches: files cached under a value now refused hold nearest-neighbour data. Nothing reads them again, and only dft_cache_clear() reclaims them. NEWS says so.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PRhUJsuKABLfpBGktPoiBN