docs(benchmarks): condense CAMPAIGN.md's finished checklist rows and history - #363

Closes #360. Docs only: benchmarks/experiments/CAMPAIGN.md goes from 121,855 to 85,667 bytes.

What was cut. In the section 7 re-pin checklist, every built, done, decided or fixed row is condensed to its number, one sentence on what changed and the pointer it already has (commit, file, test, DECISIONS and BUGS citations); the implementation detail they restated lives in the code, tests and the cited entries. Rows that are open, or that carry an instruction for a future run (2, 6, 7, 9, 10, 11, 15, 17, 19, 20, 22, 30 to 36, 76, 77, 81, 84, 87 and the owed halves or warnings of 12, 21, 60, 72, 73, 74), stay whole or keep their instruction. The "Why Phase B is not parallel" and "Why the build/query phase split is not adopted" sections are two lines each, the "ONE LINE OF DEVELOPMENT" history is its rule, section 8 is its numbered steps without the dated owner quotes and stories, the "last swept" and "started" dates are gone, and the NumPy detail under row 8 is shorter. The mentions of upstream_ledger.py and upstream_watch.py, which are private notes tools, are removed.

No row was renumbered, reordered or deleted: rows 1 to 87 are in the same order, and no new number was typed (every numeric token in the new text is in the old text). The **The October chain on mini paragraph and the qX -> qY chain line are untouched, and rows 70 and 73 keep the phrases test_l4_analyze.py and test_jpype_pin.py read. The overrides.py in row 87 is the old name of the registry that the row itself describes, so it stays.

Checks, run with nice -n 19 taskset -c 12-15 from benchmarks/experiments after the last edit and again after rebasing on main. structure_check.py passes (gates, stage chain, ingest prose, citations), with BENCH_ENGINE_COMMIT set and the local-only results/e4decomp_* directories linked in, as it needs them. pytest on test_f10_restricted_arm.py, test_overrides.py, test_degenerate_check.py, test_docs_top_parts_refusal.py, test_jpype_pin.py, test_index_kinds.py, test_l4_analyze.py, test_ingest_prose.py, test_ts_compaction.py and test_fairness_headers.py: 292 passed. uvx pre-commit run --files benchmarks/experiments/CAMPAIGN.md passes.

🤖 Generated with Claude Code