Grammar rules should be unique

[v-jkegler]

Currently there is no mechanism in llguidance for determining a grammar rule (aka node) is unique. This

1. makes the algorithm for counted sequences sub-optimal, and may prevent it from reducing the number of rules to O(n) in the length of the sequence;
2. makes our definition of grammar unnecessarily inconsistent with the usual one on the world of parsing theory, which
3. is likely, as we add parsing features, to cause other gotchas.

[mmoskal]

This can be done in GrammarBuilder; I guess we should also try to do it in Grammar::optimize

[hudson-ai]

@mmoskal linked issue for hash consing?

[mmoskal]

I think this would be for both hash-consing in the GrammarBuilder and also Hash of (node_id, num_repetitions) cache in repeat_exact and (a separate but same shape) in at_most

[v-jkegler]

See Commit 5bc30a8

[hudson-ai]

@v-jkegler uniqueness is enforced a bit further up the stack when we are building a "TopLevelGrammar", see my recent commit here
https://github.com/microsoft/llguidance/blob/29655f3aef74a21fcdbfa7c4e349c07baa89e62d/parser/src/grammar_builder.rs#L186