add in-place Tree::drop impl

[tjjfvi]

Closes #118

[HigherOrderBot]

Perf run for aac7d9b:

file              mode        main          aac7d9b736ef
========================================================
merge_sort        intr-singl       7.481 s       7.269 s
                  intr-multi       3.937 s       3.696 s
--------------------------------------------------------
boom              intr-singl       2.242 s       2.246 s
                  intr-multi       2.514 s       2.917 s
--------------------------------------------------------
bitonic_sort_lam  intr-singl      11.209 s      11.294 s
                  intr-multi       5.701 s       5.478 s
--------------------------------------------------------
radix_sort_lam    intr-singl       9.550 s       9.659 s
                  intr-multi       4.907 s       4.903 s
--------------------------------------------------------
sum_tree          intr-singl       9.163 s       9.403 s
                  intr-multi       4.359 s       4.476 s
--------------------------------------------------------
c2                intr-singl       0.000 s       0.000 s
                  intr-multi       0.001 s       0.001 s
--------------------------------------------------------
sum_tail          intr-singl       1.140 s       1.134 s
                  intr-multi       1.172 s       1.297 s
--------------------------------------------------------
sum_rec           intr-singl      14.458 s      14.272 s
                  intr-multi       7.393 s       7.432 s

[developedby]

This is quite clever, did you just think of it or did you reference something

[tjjfvi]

I've been aware of the general idea of this approach for a while; I think I first saw it here: https://matklad.github.io/2022/11/18/if-a-tree-falls-in-a-forest-does-it-overflow-the-stack.html, but that only describes implementing it for binary or variable-width trees, not mixed-width trees like here.

(I'd imagine this approach could also be used in hvm-lang.)