A comptime_int-bounded alloca would achieve those goals, plus would be more space-efficient on average than the current strategy of always pessimistically allocating for the worst case scenario.

  @alloca(T: type, count: usize, upper_bound_count: comptime_int)

with the added bonus that if `count` is small, you can avoid splitting the stack around a big chunk of unused bytes. Don't underestimate the important of memory locality on modern CPUs.