Try out making a better page cache

[diondokter]

Queue fuzzer is not happy yet

[diondokter]

Thanks for the review @Wassasin! In the end I went with a different approach, one that I can actually understand instead of something that works incidentally. That approach is now also documented a little bit. It doesn't have perfect knowledge, but the performance is still very good (see the changed perf stats in the tests).

What I don't like is that all the calculations are now taking up .text space...
So maybe I should also keep around the old impl and have people choose. The old impl is smaller and faster and will consume less ram when you only have ~8-16 pages or so.

[github-actions]

PR build size

  text	   data	    bss	    dec	    hex	filename
 18704	     56	   1728	  20488	   5008	example

Diff

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +8.8% +48.6Ki  [ = ]       0    .debug_str
   +17% +3.81Ki  [ = ]       0    .strtab
  +0.2% +1.30Ki  [ = ]       0    .debug_info
  +0.3%    +365  [ = ]       0    .debug_line
  +1.2%    +224  +1.2%    +224    .text
  +2.3%    +192  [ = ]       0    .symtab
  +0.3%    +168  [ = ]       0    .debug_ranges
  +0.3%    +120  [ = ]       0    .debug_frame
  +0.6%    +104  [ = ]       0    .debug_aranges
   +19%      +7  [ = ]       0    [Unmapped]
  -0.1%      -5  [ = ]       0    .debug_abbrev
  -0.0%     -61  [ = ]       0    .debug_loc
  +3.3% +54.8Ki  +1.1%    +224    TOTAL

[diondokter]

Todo: Update readme