Make the quarantine partition disjoint (111 + 5 + 313 = 429), and correct the run id for the eight that tipped

[briansrls]

Summary

src/v2/workflow/floor_expected_red.dag's declared rung-drop block stated a
population that did not partition. It read 111 enrolled, 111 over budget, 207
healthy — which sums to 429 only by counting the 102 rows that are BOTH enrolled
here AND budget-interrupted on both sides of the sum. Enrolment and cost are
independent axes; the earlier text treated them as one.

This corrects the block to a disjoint partition taken on enrolment, with the cost
facts stated inside each part:

  111  enrolled here and removed (the 469 -> 358 delta)
         = 9 held red + 97 budget-refused + 5 passed-over-budget,
           which run 32177951514 partitions exactly (469 = 367 + 97 + 5)
    5  unenrolled and budget-refused
  313  passing when relocated, enrolled nowhere, counted by no number in this file
         of which 9 later tipped (8 + 1 straggler) -> 304 never observed over budget
  ---
  429  = grep -c '^test fn' over the 38 relocated modules, independently

It also fixes a wrong run id that survived three rewrites of this block, including
mine: the eight that tipped were observed in 32186963581, not 32189985063. The
run counters decide it without reading any prose — 32186963581 failed=8 known_red_held=361, 32189985063 failed=1 known_red_held=358, 32193032348 failed=1 known_red_held=358. Only one run observed eight. The SECOND WAVE paragraph
lower in the same file named the right one the whole time: one file, two accounts of
one population, which is why the counters and not the neighbouring sentence are the
authority here.

And it separates three figures that were circulating as if they were one, because
each names a different set and none may be written as a bare budget_refused:
102 budget-interrupted globally · 107 over-budget-or-failing at quarantine time
(102 + 5 passed-over-budget) · 116 including the 9 that tipped afterwards.

Why this matters beyond the file

Per DESIGN §4b, a rung drop must declare a bounded population, and per the
floor_expected_red shrink-monotonicity thread, sharp-raven-273's monotone-shrink
wall reads this carrier as its baseline. A stale or non-disjoint count here is
load-bearing for work that is not mine. A bounded population is the one thing a rung
drop must not round.

Attribution

The 358/111 corrections are tidy-lark-471's external review, not mine. The
disjointness defect and the run-id correction are mine, and both were found by
re-deriving the partition rather than re-reading the prose.

Test plan

• Comment-only change, verified mechanically, not asserted:
  git diff origin/main...HEAD -- <file> | grep -E '^[+-]' | grep -v '^[+-][[:space:]]*//'
  returns empty — no non-comment line is added or removed. 134 insertions,
  14 deletions, all //.
• Block placement is a standalone leading // block on a module-scope declaration
  (DESIGN §4c admits exactly this form; trailing and in-body forms refuse).
• No local execution. There is no target/release/gunbc in this worktree, and a
  whole-tree build or floor run is a CI job rather than a session job under the fleet
  resource policy. Parse and floor verification are relied on from CI on this PR.
• Every number above is checkable against the tree (grep -c '^test fn' over the 38
  relocated modules) and against the named run counters. None is copied from another
  session's measurement.

Work item: fix floor over budget tests (dashboard work item — no GitHub issue to close).