Auto merge of #75257 - ssomers:btree_74762_again, r=Mark-Simulacrum
BTreeMap: better way to postpone root access in DrainFilter
A slightly more elegant (in my opinion) adaptation of #74762. Benchmarks seem irrationally pleased to:
```
benchcmp old new --threshold 5
name old ns/iter new ns/iter diff ns/iter diff % speedup
btree::map::clone_fat_100_and_remove_all 215,182 185,052 -30,130 -14.00% x 1.16
btree::map::clone_fat_100_and_remove_half 139,667 127,945 -11,722 -8.39% x 1.09
btree::map::clone_fat_val_100_and_remove_all 96,755 81,279 -15,476 -16.00% x 1.19
btree::map::clone_fat_val_100_and_remove_half 64,678 56,911 -7,767 -12.01% x 1.14
btree::map::find_rand_100 18 17 -1 -5.56% x 1.06
btree::map::first_and_last_0 33 35 2 6.06% x 0.94
btree::map::first_and_last_100 40 54 14 35.00% x 0.74
btree::map::insert_rand_100 45 42 -3 -6.67% x 1.07
btree::map::insert_rand_10_000 45 41 -4 -8.89% x 1.10
btree::map::iter_0 2,010 1,759 -251 -12.49% x 1.14
btree::map::iter_100 3,514 2,764 -750 -21.34% x 1.27
btree::map::iter_10k 4,018 3,768 -250 -6.22% x 1.07
btree::map::range_unbounded_unbounded 37,269 28,929 -8,340 -22.38% x 1.29
btree::map::range_unbounded_vs_iter 31,518 28,814 -2,704 -8.58% x 1.09
```
r? @Mark-Simulacrum
d19d7e2755