perf: optimize k-mer pipeline with compile-time tables

This commit shifts entropy and lookup table generation to compile time via a new build script, eliminating runtime overhead. It replaces heap-allocated queues in rolling statistics with a stack-allocated, const-generic ring buffer for cache-friendly operations, and implements `size_hint` on `SuperKmerIter` for efficient iterator consumption. Additionally, it establishes the baseline profile configuration and sets global k-mer parameters.

src/obikseq/src/kmer.rs

@@ -118,7 +118,7 @@ pub struct KmerOf<L: KmerLength>(u64, PhantomData<L>);
 impl<L: KmerLength> KmerOf<L> {
     /// Wrap a raw left-aligned u64 value as a kmer.
     #[inline]
-    pub fn from_raw(raw: u64) -> Self {
+    pub const fn from_raw(raw: u64) -> Self {
         KmerOf(raw, PhantomData)
     }