fix: recompute overlap after trimming for merge mode (#675)

The overlap caching optimization introduced in v1.3 computed the overlap
result once before adapter/polyX/maxLen trimming and reused it for the
merge step. This caused the merge to use stale overlap data from
untrimmed reads, leading to pairs that should merge post-trim being
incorrectly routed to the unpaired output path instead.

Always recompute the overlap on the post-trim reads before merging,
restoring v1.2 behavior where the merge step analyzed trimmed reads.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: KimBioInfoStudio

src/peprocessor.cpp

@@ -513,10 +513,9 @@ bool PairEndProcessor::processPairEnd(ReadPack* leftPack, ReadPack* rightPack, T
         // merging mode
         bool mergeProcessed = false;
         if(mOptions->merge.enabled && r1 && r2) {
-            if(!ovComputed) {
-                ov = OverlapAnalysis::analyze(r1, r2, mOptions->overlapDiffLimit, mOptions->overlapRequire, mOptions->overlapDiffPercentLimit/100.0);
-                ovComputed = true;
-            }
+            // Always recompute overlap on post-trim reads for merge (fixes #675)
+            ov = OverlapAnalysis::analyze(r1, r2, mOptions->overlapDiffLimit, mOptions->overlapRequire, mOptions->overlapDiffPercentLimit/100.0);
+            ovComputed = true;
             if(ov.overlapped) {
                 merged = OverlapAnalysis::merge(r1, r2, ov);
                 int result = mFilter->passFilter(merged);