LLVM tier: adjust num of IC slots depending on the num of times the IC got rewritten in the bjit tier

[undingen]

I chose IC slot size heuristic just by running a few times the benchmarks and taking the best one, we may want to adjust it later.

           django_template3.py             2.2s (6)             2.2s (2)  -0.5%
                 pyxl_bench.py             1.9s (6)             1.8s (2)  -3.8%
     sqlalchemy_imperative2.py             2.3s (6)             2.3s (2)  -1.3%
                pyxl_bench2.py             1.5s (6)             1.3s (2)  -14.5%
       django_template3_10x.py            14.3s (6)            14.2s (2)  -0.5%
             pyxl_bench_10x.py            14.7s (6)            13.9s (2)  -5.4%
 sqlalchemy_imperative2_10x.py            18.1s (6)            17.4s (2)  -3.4%
            pyxl_bench2_10x.py            13.8s (6)            11.6s (2)  -15.9%
                       geomean                 5.4s                 5.1s  -5.9%

[kmod]

Wow, 15% (even on pyxl_bench2) is pretty impressive :)