Reapplying the Decay Range and other opts

[Matthew Parkinson (mjp41)]

This reapplies the DecayRange PR to the latest version of snmalloc. It also

• removes some unrequired initialisation that occurs in a fast path for large allocations. Finally, it adds a LargeObjectCache to improve performance of large allocations.

This is a very draft PR with a lot of reviewing of AI refactorings required.

This might increase footprint considerably, and care is needed to configure who it operates precisely.

[akrieger]

If I understand correctly, Windows will only faults in touched pages and not the entire range. I don't know the behavior for other kernels but that's what the documention I can find says https://learn.microsoft.com/en-us/windows/win32/memory/reserving-and-committing-memory "... As an alternative to dynamic allocation, the process can simply commit the entire region instead of only reserving it. Both methods result in the same physical memory usage because committed pages do not consume any physical storage until they are first accessed."

[Matthew Parkinson (mjp41)]

Yeah the benchmark is not going to fault all the pages in. Just the first and last, and this would be true of most platforms. If this lands, I'll fix. Thanks

[Matthew Parkinson (mjp41)]

I think this is ready to review. But we should address #814 before we commit this, as it may either hide or worsen this behaviour.