perf(propagate): don't increase the stack when there is only one subscriber

[johnsoncodehk]

Benchmark Results

main branch

benchmark	avg	min	p75	p99	max
propagate: 1 * 1	1.59 µs/iter	1.56 µs	1.61 µs	1.65 µs	1.67 µs
propagate: 10 * 1	14.32 µs/iter	14.23 µs	14.34 µs	14.40 µs	14.47 µs
propagate: 100 * 1	139.54 µs/iter	136.83 µs	138.00 µs	182.58 µs	307.58 µs
propagate: 1 * 10	8.61 µs/iter	8.59 µs	8.62 µs	8.63 µs	8.63 µs
propagate: 10 * 10	84.50 µs/iter	82.92 µs	83.58 µs	109.29 µs	249.54 µs
propagate: 100 * 10	843.57 µs/iter	831.71 µs	837.38 µs	983.25 µs	995.58 µs
propagate: 1 * 100	77.82 µs/iter	76.08 µs	76.96 µs	102.38 µs	241.33 µs
propagate: 10 * 100	767.59 µs/iter	754.83 µs	761.71 µs	913.63 µs	938.08 µs
propagate: 100 * 100	7.67 ms/iter	7.57 ms	7.73 ms	7.80 ms	7.80 ms

This PR

benchmark	avg	min	p75	p99	max
propagate: 1 * 1	1.52 µs/iter	1.50 µs	1.54 µs	1.55 µs	1.56 µs
propagate: 10 * 1	13.89 µs/iter	13.86 µs	13.91 µs	13.92 µs	13.92 µs
propagate: 100 * 1	137.14 µs/iter	134.75 µs	135.92 µs	174.79 µs	296.29 µs
propagate: 1 * 10	8.07 µs/iter	8.06 µs	8.07 µs	8.08 µs	8.09 µs
propagate: 10 * 10	79.17 µs/iter	77.79 µs	78.42 µs	90.38 µs	231.33 µs
propagate: 100 * 10	786.38 µs/iter	775.50 µs	781.67 µs	926.71 µs	951.96 µs
propagate: 1 * 100	72.09 µs/iter	70.67 µs	71.46 µs	82.42 µs	244.08 µs
propagate: 10 * 100	717.08 µs/iter	706.29 µs	712.83 µs	861.29 µs	888.17 µs
propagate: 100 * 100	7.15 ms/iter	7.07 ms	7.21 ms	7.25 ms	7.26 ms

[johnsoncodehk]

This is a fast path that can only be achieved without using recursion propagate call.

Before this PR, propagate always used stack to fall back to the previous node.

In this PR, propagate will only roll back to the previous branch.