Fix pathfinding

[glebm]

The previous implementation didn't behave quite like A-* is supposed to.

After trying to figure out what's causing it and giving up, I've reimplemented it in a straightforward manner.
Now it seems to work a lot better.

Also increases maximum player path length to 100 steps.
We still only store the first 25 steps in the save file for vanilla compatibility.

Why draft?

1. Need to check how much smarter monsters are with this.
2. New demo needs to be recorded.

Benchmarking shows that the new algorithm is about twice as fast:

Comparing build-reld-path-bench/path_benchmark to build-reld/path_benchmark
Benchmark                              Time             CPU      Time Old      Time New       CPU Old       CPU New
-------------------------------------------------------------------------------------------------------------------
BM_SinglePath_pvalue                 0.0002          0.0002      U Test, Repetitions: 10 vs 10
BM_SinglePath_mean                  -0.3270         -0.3271          2917          1963          2917          1963
BM_SinglePath_median                -0.3247         -0.3247          2907          1963          2907          1963
BM_SinglePath_stddev                -0.8035         -0.8067            15             3            15             3
BM_SinglePath_cv                    -0.7080         -0.7127             0             0             0             0
BM_Bridges_pvalue                    0.0002          0.0002      U Test, Repetitions: 10 vs 10
BM_Bridges_mean                     -0.7641         -0.7641         59742         14095         59736         14094
BM_Bridges_median                   -0.7617         -0.7618         59083         14077         59077         14074
BM_Bridges_stddev                   -0.9810         -0.9810          2719            52          2719            52
BM_Bridges_cv                       -0.9194         -0.9195             0             0             0             0
BM_NoPath_pvalue                     0.0002          0.0002      U Test, Repetitions: 10 vs 10
BM_NoPath_mean                      -0.8406         -0.8406        127807         20371        127792         20368
BM_NoPath_median                    -0.8406         -0.8406        127556         20330        127540         20328
BM_NoPath_stddev                    -0.9138         -0.9141          1109            96          1111            95
BM_NoPath_cv                        -0.4592         -0.4609             0             0             0             0
OVERALL_GEOMEAN                     -0.4405         -0.4405             0             0             0             0

If we go ahead with this fix, we can consider increasing the max path length for the player in a follow-up PR.

Here is what a somewhat long path looks like:

We may want to tune it further to prefer diagonal steps first more reliably.

[kphoenix137]

Does this PR also fix #6103 ?

[glebm]

Probably not, this fixes the pathfinding algorithm but doesn't change how stuns are handled.

[glebm]

@ephphatha @qndel @AJenbo Ready for review and playtesting!