refactor: eliminate notify recursion

[medz]

clk: ~0.05 GHz
cpu: Apple M3 Max
runtime: node 23.9.0 (arm64-darwin)

Before:

benchmark	avg	min	p75	p99	max
propagate: 1 * 1	946.52 ns/iter	921.31 ns	951.22 ns	967.06 ns	972.19 ns
propagate: 1 * 10	4.09 µs/iter	4.03 µs	4.09 µs	4.30 µs	4.30 µs
propagate: 1 * 100	34.57 µs/iter	34.18 µs	34.63 µs	34.66 µs	34.86 µs
propagate: 10 * 1	7.20 µs/iter	7.19 µs	7.22 µs	7.22 µs	7.23 µs
propagate: 10 * 10	38.08 µs/iter	37.57 µs	38.21 µs	38.25 µs	38.26 µs
propagate: 10 * 100	346.01 µs/iter	321.71 µs	347.13 µs	368.67 µs	467.71 µs
propagate: 100 * 1	69.05 µs/iter	63.63 µs	68.46 µs	83.42 µs	89.92 µs
propagate: 100 * 10	384.42 µs/iter	359.38 µs	385.75 µs	416.67 µs	428.29 µs
propagate: 100 * 100	3.44 ms/iter	3.41 ms	3.44 ms	3.48 ms	3.49 ms

signal: 1173.70 KB
computed: 2620.52 KB
effect: 3091.95 KB
tree: 5029.80 KB

After:

benchmark	avg	min	p75	p99	max
propagate: 1 * 1	948.87 ns/iter	930.30 ns	952.74 ns	974.88 ns	977.20 ns
propagate: 1 * 10	4.07 µs/iter	4.03 µs	4.08 µs	4.11 µs	4.12 µs
propagate: 1 * 100	34.46 µs/iter	34.25 µs	34.55 µs	34.66 µs	34.73 µs
propagate: 10 * 1	7.24 µs/iter	7.20 µs	7.25 µs	7.27 µs	7.28 µs
propagate: 10 * 10	38.07 µs/iter	37.98 µs	38.12 µs	38.14 µs	38.18 µs
propagate: 10 * 100	345.86 µs/iter	322.67 µs	346.63 µs	364.63 µs	491.71 µs
propagate: 100 * 1	69.11 µs/iter	63.96 µs	69.08 µs	76.96 µs	102.71 µs
propagate: 100 * 10	381.57 µs/iter	356.21 µs	382.00 µs	399.13 µs	418.54 µs
propagate: 100 * 100	3.42 ms/iter	3.39 ms	3.43 ms	3.46 ms	3.48 ms

signal: 1173.70 KB
computed: 2620.52 KB
effect: 3091.95 KB
tree: 5026.55 KB

---

The performance change is minimal, because the error of each device run needs to be taken into account, so it is concluded that this has no impact on performance. But it will be slightly better if there are many nodes. (Still negligible)

In terms of memory usage, tree has been reduced.

[johnsoncodehk]

Since the notify function isn't a performance hotspot, I think keeping the Surface API code simple and easier for people to understand will have a greater benefit.

Thanks for the PR!