Refactor similarity engine and implement parallel clustering

Introduces parallel processing for clustering, refines scoring mechanisms, and expands test coverage for the alignment functionality.

This change introduces parallel processing capabilities within the clustering logic, updates the core similarity engine with new data structures and scoring methods, and implements a comprehensive test suite for the alignment tool. It also adds new command-line options for configuring alignment parameters and reference lengths.

pkg/obialign/lcsmin.go

@@ -62,8 +62,9 @@ func bandedLCSAtLeast(a, b []byte, minLCS int, ws *bandedLCSWorkspace, useSIMD b
 	}
 
 	// The score alone is much cheaper than the score with the length of the
-	// alignment: the latter is only computed when the pair reaches minLCS.
-	if lcsScoreSIMD(a, b, kmin, kmax, minLCS, ws) < minLCS {
+	// alignment: the latter is only computed when the pair reaches minLCS. With
+	// no minimum there is nothing to reject, and the pass would be wasted.
+	if minLCS > 0 && lcsScoreSIMD(a, b, kmin, kmax, minLCS, ws) < minLCS {
 		return -1, -1
 	}
 	encodeBases(ws, a, b)