[ayiskakov]

What

Tools/upstream-candidates.py lists the upstream commits this fork has not taken yet and says which of them apply, so upstream's fixes (the BLE stack above all) can be taken one at a time with git cherry-pick -x instead of by a wholesale merge, which stopped working once Android, Oura/Polar/Huami and WHOOP 4.0 were removed here (three days of upstream came to 84 conflicts, 72 in deleted files).

• Hidden: commits that only touch code this fork removed; commits already taken (a -x trailer, or the same patch id); commits declined in Tools/upstream-skip.txt. --all lists them with the reason.
• For the rest: a git merge-tree simulation against HEAD (no index or work-tree change), reported as clean, clean, N rm (also edits N files this fork removed) or N conflicts. A commit that only conflicts because it builds on an earlier candidate is re-simulated on top of it and marked ^ when it then applies.
• --apply <sha>...: picks in upstream order with -x, drops the files this fork removed when those are the only conflicts, and stops before any commit that needs a hand resolution, leaving the tree clean. A file missing only because the upstream commit that adds it has not been taken is a conflict, never a drop.

Tools/upstream-skip.txt starts with today's review: six commits that do nothing for a 5.0/MG Apple build or that this fork has superseded, plus upstream's testing build numbers by subject.

Tools/upstream-candidates.py --fetch                 # what is new
Tools/upstream-candidates.py --area ble,protocol -v  # files, conflicts, what a ^ needs
Tools/upstream-candidates.py --apply <sha>...        # on a branch, not main

Verification

• Tools/test_upstream_candidates.py: 40 tests against a throwaway repository shaped like this history (Android-only commits, both-platform fixes, clashes with the fork, chains, a file added by an earlier upstream commit, picked, same patch, declined). Isolated from the user's git config.
• Mutation checks: reverting the directory rule, the pre-simulation stop in --apply, or the removed-versus-not-yet-added rule each fails its tests.
• Real repo: all 16 commits hidden as removed-only were checked by hand (every path is in android/, Oura or the parity files); --apply of five real upstream commits on a throwaway worktree landed them in upstream order with -x trailers, no android/ resurrected, and an out-of-order follow-up applied once its predecessor was in.
• Full core Tools suite locally: 106 tests; the 4 failures are in test_check_source_references (macOS /var vs /private/var temp paths), unrelated and unchanged by this PR.