fix: loop indicator token unstable for Rule 1 about turing HOT 4 CLOSED

Going to extra lengths to indicate the final index, which it is having difficulty doing for the Rule 1 case especially.

TAPE 0░ 1░ 2░ 3█ 4░ 5░ 6░ 7█ 8█ 9█
PRINT 1/9 ░ ░ ░ █ ░ ░ ░ █ █ █
INDEX 1=9 DONE
PRINT 0/9   ░ █ ░ ░ ░ █ ░ ░ ░ ░
PRINT 1/9   ░ ░ ░ █ ░ ░ ░ █ █ █

It's tripping so hard that it's willing to write 1=9 before thinking to continue looping.

from turing.

I may cheat, and simply add Rule 1 to the training set, and whatever it is missing it should learn, but it would be more elegant for it to learn it without cheating.

I'm thinking it might be the proximity to the other PRINT statement.

from turing.

It weirdly had just never seen a 1 != 9 comparison before, since no training examples had 9 steps, and for the Rule 1 case, it would for, whatever reason, think one is equal to nine.

Adding a single case for the number of steps we plan to test solved this, since it had seen 1/9 not complete. I will add two to be safe.

from turing.

The majority of the issues that needed extra labeling for the index seem to have been caused by it having no 10-step (9-max-index) examples. I added several and tightened up the tape syntax, which is like 5-10% lighter.

from turing.