I'd argue the main reason Zig/C programmers use arenas is for correctness, not performance. You might think of arenas as a performance thing if you consider the alternative to be a GC, but the alternative in Zig/C is usually to do things manually.
Like the other commenter said, “You use an arena when you have computation that might consume a bunch of scratch memory, and then you want to free all of that memory at once when the computation is done.” That’s the entire point of arenas.
"Arena" normally also implies the allocation algorithm is stack-like, but this is not a hard requirement tied to the implementation.
In any case the quality that allows to quickly free all tied memory is not exclusive to a bump-type allocator; any allocator that exists on its own can do that. It is just that we do not often see standalone malloc-like allocators. But they are totally possible. A bump allocator supports other actions that are indeed exclusive to its design: partial free (set a mark and later "return" to it and free everything that was allocated after that mark) or allocating the last chunk step-by-step ("growing"). These are indeed unique to bump allocator, but the way it frees all memory is not.
For example, one might want to combine a bump and slab allocator: generally bump, but allow for 'free' for small sizes with subsequent reuse of the freed slab. This would be a tad slower than pure bump, but will have a better use of memory in certain scenarios.