Allows playing with the shortcuts more easily, while keeping a strictly
exhaustive strategy around for comparison.

If we have enough robots to collect enough materials every turn to build
a new geode robot every turn, and enough materials to build a geode
robot this turn, it's trivial to calculate how many geodes it's possible
to collect in the remaining time.

Because it can never be negative, and some of the blueprint combinations
can cause it to wrap. This gives us one more factor of two breathing
room.

Either we don't need it, or we did but it's too late.

Speeds up part 2 from 570s (9m30s) to 5.2s

When the number of iterations was small, looking near the original
position of the num was a good idea. As the number of iterations
increases, the positions are fairly random, so speed of lookup is mostly
a factor of how quickly we can scan the whole buffer.

Speeds up part 2 of the problem from 2.2s to 0.12s.

We now manage the general case of dividing up the work between as many
plumbers as available, as equally as possible. Works for the test case
and the full input case.

(Actually, it seems to speed up the full input case, from around 68s to
55s!)

By realising that the first time we hit the need to skip, the items to
be skipped will be in sorted order; to change them to reverse-sorted
order just requires reversing the items, not re-sorting them backwards.

(On part 1 of the puzzle with a 40 minute limit, the timing goes from
6.6s to 4.3s, average of 5 runs.)