Improve the memory model

[booniepepper]

The memory model is incredibly naive right now: Arena allocate everything, and only free when the entire program completes. This is passable (and fast!) for simple cases, but would scale terribly for long-running processes since no memory is almost never freed.

Maya gave some good critiques in this Zig discord thread:

https://discord.com/channels/605571803288698900/1128605557796905020

[edt-xx]

Yah. This is a really good idea. duck tape is nice in both its physical and software variants. Was going to open a bug on memory usage when I saw this.... If you want your process killed, try this increasing 20 to 100000 or so:

seq 20 | ./dt 'deq swap drop [ deq swap \d: [ d divisor? not ] filter d pl ] [ swap dup rot len ] while'

ps. Its a little pita to use 'swap dup rot ' instead of over ( a b --- a b a )...

[booniepepper]

For memory management, I'm working through exactly what I want to do in this project: https://github.com/booniepepper/dt-rewrite

Getting close to a point where I'll start incorporating it