refactor: remove equivalence class folding from entropy pipeline

Removes circular-reverse complement machinery and explicit k-mer canonicalization across the entropy pipeline. Frequency tallying and Shannon entropy computation now operate directly on raw k-mer values, eliminating prior score inflation and alignment-dependent artifacts while preserving orientation invariance. Updates build scripts to generate normalized lookup tables for k-mer lengths 1–6, restricts the public API to `EntropyTracker`, and bumps crate versions. Documentation is updated to reflect the simplified raw-value approach and revised module structure.

src/obiskbuilder/src/rolling_stat.rs

@@ -1,4 +1,4 @@
-use obikentropy::{EntropyTracker, sub_word_canon};
+use obikentropy::EntropyTracker;
 use obikseq::kmer::{Minimizer, hash_kmer};
 use obikseq::params;
 
@@ -136,8 +136,6 @@ impl RollingStat {
         self.rolling_rck =
             ((self.rolling_rck >> 2) | ((cnuc as u64) << ((k - 1) * 2))) & self.k_mask;
 
-        let canon = sub_word_canon(self.rolling_k);
-
         self.received += 1;
 
         if self.received >= m {
@@ -170,7 +168,7 @@ impl RollingStat {
             }
         }
 
-        self.entropy.push(self.received, canon);
+        self.entropy.push(self.received, self.rolling_k);
     }
 
     pub fn ready(&self) -> bool {