Two lines where 52,000 were, and the resume that wrote one twice

Fixed TMR with one good replica and one stuck one detects a disagreement on every tick, and wrote one line per tick: 1.2 MB in a 110 second run, with the package due that day. The event writer now keeps a streak: what, detail, first tick, last tick, count. Identical consecutive events collapse to the first line and a closing one, 4500,detected,Compare repeated 4250x ticks 251-4500. A run that wrote 52,000 lines writes two. The TMR log fell to 262 bytes. Version 0.3.0 was cut for it, so the package could use a released binary.

The next day CI saw killed,0 twice in one log. Once, under emulation. The voter flushed the log before it wrote the checkpoint, and a reset between the two left events in the log that the resumed run wrote again. The checkpoint now records how many bytes of the log belong to it, and the streak in progress. On resume the log is truncated to that length and the streak continues, so the closing line after a resume still counts from the start. Scenario 18 plants a killed line after the checkpoint tick and checks that the resume removes it.

The scenario had a defect of its own. grep -c exits 1 when it counts zero, and under set -e that was the end of the script.