Give up on an engine only when it has never run on this machine

[tamnd]

The 32 core box stopped at 45 cells of 60 after #81 was deployed to it, and its own sweep log says why.

bench_yo-threads_16-pipeline_10-perf_no-run_2.json failed: on the sets pass the first of 16 load generator threads finished after 3.761 seconds where the last took 31.477
yo has failed 3 times in a row, so the rest of its cells are being left rather than failing one at a time for the next day
swept 60 cells in 3m 19s: 0 measured here, 45 already on disk, 3 failed, 11 left alone because their engine was given up on, 1 left alone after 4 tries each

GIVE_UP fires before TRIES ever can. The five runs of one shape are consecutive in the matrix order, so a shape that box cannot measure is three failures in a row on the third run, the engine goes down, and the eleven cells after it are never attempted. The count from #81 needs four full sweeps to bite and the engine is gone after one.

The rule was written for an engine the machine cannot run at all, where attempting a thousand cells to learn that costs a day. That is a real case and it stays. What tells the two apart is whether the engine has ever produced a cell here. One on the disk, or one measured in this session, and the failures are about a shape rather than about the engine, so the sweep walks on. The cost of walking on is one run for each remaining bad cell, once, because the attempt count then leaves those cells alone on the next sweep. The cost of not walking on is a sweep that stops at the same place however many times it is restarted.

put_down is the whole decision and it has its own test. whats_left collects the engines with a cell already on the disk, which is what makes this work on a restart rather than only within one session.