test_drop_file_cache: posix_fadvise cache-drop proof fails when /tmp is tmpfs

Row: -

python3 -m unittest tests.tools.test_drop_file_cache fails on a clean,
unmodified main (bcade48d6) on this box:

ERROR: test_posix_fadvise_evicts_all_inventoried_resident_pages
tools.bench.drop_file_cache.CacheDropError: cache-drop proof retained 6291456 resident bytes

Reproduced with no other change in the tree; not caused by any pending work.
This box's /tmp is a 32G tmpfs mount (df -h /tmp). posix_fadvise(..., POSIX_FADV_DONTNEED) on a tmpfs-backed file cannot evict pages the way it
does on a real block-device-backed file, because tmpfs pages are the storage,
not a cache of it -- so the eviction proof this test asserts on can never
succeed there. Either the test needs a non-tmpfs scratch directory, or
drop_file_cache needs to detect a tmpfs target and skip/report accordingly.

Filed without a fix in the same flow: found as a side effect of unrelated
work (ORACLE-VLLM-ROCM-GFX1200-DOCKER, #2961) and needs its own
investigation into the right fix (relocate the test's scratch dir vs. teach
drop_file_cache about tmpfs) rather than a guess landed in a hurry.

Start by running python3 -m unittest tests.tools.test_drop_file_cache and inspect tests.tools.test_drop_file_cache plus tools.bench.drop_file_cache. Reproduce the failure with /tmp on tmpfs, then determine whether the scratch directory or cache-drop handling should change. Done means the test has defined behavior on tmpfs without weakening the proof on block-device-backed storage.