Explore prefetch-compatible trie layout

[eiennohito]

When building a dictionary, we need to build Trie index.

Probably, we shouldn't move away from darts-clone-compatible structure, but there can be variations on the construction. It would be interesting to explore whether it is possible to make prefetch-compatible trie that could load the trie entries in L1 cache for the next symbol speculatively, even with not 100% accuracy (e.g. getting hits for high-frequency entries).

Rust Implementations to Note: https://github.com/takuyaa/yada (it seems to follow darts-clone internal structure with original darts construction procedure)