Corgi join: consolidate compound values when advancing reorders product times

[frankmcsherry]

SideScratch::stage_runs (interactive/src/corgi/join.rs) has a single-chunk path that assumes joining each time with the lower bound keeps a value's times in ascending order ("join is monotone"), so equal (coordinate, time) entries stay adjacent and adjacent consolidation suffices.

For partially ordered times that is false. With product times (0,2), (1,0), (1,2) and lower bound (1,0), the advanced times are (1,2), (1,0), (1,2). With diffs +1, +1, -1 for one compound value, the +1/-1 pair at (1,2) is not adjacent, so it survives staging, and the staged entries are no longer sorted:

left:  [(0, (1, 2), 1), (0, (1, 0), 1), (0, (1, 2), -1)]
right: [(0, (1, 0), 1)]

The fix checks for an inversion within a value coordinate after staging and consolidates only then; ordered runs keep the existing path. The primitive-value path already consolidated, so this covers compound values. +23 −3, one file.

• New test compound_tokens_consolidate_after_product_time_reordering; it fails on master-next with the output above.
• cargo test --release -p interactive: 104 passed, 5 ignored.
• Found while profiling worldgen on DDIR (Astra). On that workload, with the fix, 27/27 tour steps and 13/13 warm-move reference checks agree with the native engine at 1 and 4 workers, and timing is flat (+0.2% / −0.2%).

🤖 Generated with Claude Code