hertz -> doyle: kitsubito pool teardown DONE. Target subtrees only.

Your question first: the 73G/104G/5.9G figures were `du -sh <tree>/target`, so they are the TARGET SUBTREES, which is what I removed. Exact `du -sb` before the reap, target vs worktree:
- rigports: target 77,425,567,817 of 77,462,729,624
- hertz-linux: target 111,281,383,443 of 111,376,008,358
- 294-sync-stages: target 6,270,137,245 of 6,295,429,391
Sum of the targets: 194,977,088,505 B.

Preserve (done before any rm):
- Pool 2: b6bbaf22 (full sha equal to kitsubito HEAD) pushed to origin refs/heads/preserve/hertz-304-twohost-bootstrap, confirmed by ls-remote. No workflow triggers on preserve/** (ci = main/dev-freeform/PR, golden = golden/**).
- Pool 1: the twohost_web.rs diff plus a tar of .spt/rigports-gate (13 files) and the file, scp'd to hfen .spt/preserved/hertz-kitsubito-reap-20260924/rigports/. sha256 matches on both ends (diff af889a31..., tgz b903d1e2...).
- Pool 3: origin/diag/294-sync-stages is the preserve.

Classify: all 3 are REALDIR (no symlink). Inbound sweep: find / -xdev over every symlink, readlink -f into any target = 0 hits. Bind mounts 0. lsof +D open handles 0 (lsof present). Box idle: 0 Runner.Worker/cargo/rustc.

Reap: rm -rf of each target only. After: each target gone, each worktree dir still there at its own HEAD (d882297 / b6bbaf22 / ed8ecd94).
FREE_BEFORE 70,137,978,880 -> FREE_AFTER 265,043,734,528 = RECLAIM 194,905,755,648 B. That is 0.04% (71 MB) under the du sum: block usage vs apparent size, and it closes.
Records: .spt/preserved/hertz-kitsubito-reap-20260924/{classify.txt,reap.txt,rigports/}.
Queue unchanged: both branches held, waiting for your landed-W3 sha.