[tannevaled]

This PR corrects docs/performance-campaign.md (#26) after reading go-geometry/predicates#4, go-sparse/sparse#10 and special's full perf table:

• predicates: the gap is 1.6–3.3× on random records and up to 6.9× in the exact stage D. The page said 1.4–3.6×. The leads from chore(deps): update actions/checkout action to v7 #4 are now on the page.
• sparse: besides the refactor (11–21 %), the solve is up to 13 % slower, and AMD and the symbolic phase allocate per node. These are the causes behind optimize's ADMM setup at 2.2–4.9× OSQP's.
• special: the single-core range is 0.07×–0.70× on all 31 rows. The page said 0.19×–0.40× because only the first rows had been read.
• Issues: lists the issues filed from the page (dd: one double-double package for the stack (stats/internal/dd duplicates special/dd) go-numerics/special#32, perf: the single-core Go loop is 0.07x-0.70x SciPy's speed on all 31 measured functions go-numerics/special#33, perf: docs/perf.md compares multicore Go with single-threaded NumPy 2.2.4; re-measure on one core against the pinned NumPy go-ndarray/ndarray#93, W0: go-scikit/refbench, the shared reference-benchmark harness #27).

🤖 Generated with Claude Code