The merge queue remembers clashes: two pulls that are green alone but red together are never batched together again until one lands #4235

Follow-up to #4053 / #4058 (parallel bisection shipped 07:35). Probe 2026-10-21 09:45Z: in the last hour the same pulls kept riding red batches - #4119 and #4120 in 5 red batches each, #4121 in 4, #4157 in 3 - sibling pulls from one builder touching /explore, each green alone; earlier today #3413 rode 10 red batches. Every red batch also cut off a parallel batch of fresh pulls.

Done:

1. When bisection isolates a red, record what it found: either a pull red on its own merge (→ back to its builder with the failing test, out of the queue), or a pair red together (→ a clash record).
2. A clash pair is never in the same candidate again until one of them lands or changes head; the second waits for the first to land, then is re-checked on the new main.
3. After a red is resolved, the remaining good pulls re-batch together immediately (no single-pull drip).
4. The tick line shows clashes=N; /boards shows red-batch rate.

Test: three pulls where A+B clash, C is fine → A and C land in one gate, then B is re-checked and lands; A and B are never gated together twice.

Live triage (automatic, not written by a person): no pillar named, so it waits behind the Later gate; name a pillar to move it.