Improve performance of TrieNode serialization

[jsdanielh]

Improve performance of TrieNode serialization.

Pull request checklist

• All tests pass. The project builds and runs.
• I have resolved any merge conflicts.
• I have resolved all clippy and rustfmt warnings.

[hrxi]

Looking at it.

[hrxi]

I did some changes to postcard(!) and the way we serialize hashes (thanks for the nice benchmarks btw), the result is as follows:

empty/serde             time:   [20.025 ns 20.300 ns 20.662 ns]
empty/plain             time:   [18.265 ns 18.381 ns 18.567 ns]

root/serde              time:   [34.466 ns 34.592 ns 34.742 ns]
root/plain              time:   [29.107 ns 29.238 ns 29.394 ns]

root_incomplete/serde   time:   [36.553 ns 36.746 ns 36.965 ns]
root_incomplete/plain   time:   [34.310 ns 34.487 ns 34.695 ns]

leaf32/serde            time:   [25.873 ns 25.951 ns 26.050 ns]
leaf32/plain            time:   [23.987 ns 24.149 ns 24.346 ns]

leaf128/serde           time:   [28.838 ns 28.896 ns 28.977 ns]
leaf128/plain           time:   [24.703 ns 24.794 ns 24.903 ns]

leaf1024/serde          time:   [35.356 ns 35.532 ns 35.784 ns]
leaf1024/plain          time:   [31.009 ns 31.204 ns 31.406 ns]

leaf16834/serde         time:   [214.04 ns 214.85 ns 215.63 ns]
leaf16834/plain         time:   [191.85 ns 192.91 ns 194.14 ns]

leaf32768/serde         time:   [680.77 ns 683.33 ns 686.13 ns]
leaf32768/plain         time:   [641.49 ns 644.56 ns 647.63 ns]

leaf131072/serde        time:   [2.9458 µs 2.9569 µs 2.9696 µs]
leaf131072/plain        time:   [2.9047 µs 2.9149 µs 2.9276 µs]

The results show that the tweaked serde/postcard serialization is consistently a bit slower than the handrolled one, but always within ~10%.

[hrxi]

Given the improved performance of the tweaked serde/postcard serialization, I think not hand-rolling our own serialization is better because it's introducing less complexity.

The way forward for this would be me pull-requesting my changes to postcard. Since we probably want to have that either way, I'm going to prepare the pull request for postcard.