fix: prevent integer overflow and truncation in MergedPileup weighted mean

The weighted mean calculation `p.total * p.left_pos / total` had two bugs:

1. Integer overflow: `p.total * p.left_pos` is Int*Int which overflows
   for large genomic positions (up to ~250M) with moderate counts.

2. Per-term truncation: integer division inside map() truncates each
   term before summing, producing inaccurate results. E.g. three
   pileups with total=1 at pos=100 with overall total=3: each term
   becomes 1*100/3=33 (truncated), sum=99 instead of 100.

Fix uses Long arithmetic for the product and performs a single division
at the end.

Author: nh13

src/main/scala/com/fulcrumgenomics/sv/tools/FilterAndMerge.scala

@@ -105,8 +105,8 @@ object MergedPileup {
     val readPairs  = pileups.sumBy(_.read_pairs)
     val counts     = NumericCounter(pileups.map(_.total).iterator)
     val total      = splitReads + readPairs
-    val leftMean   = pileups.iterator.map(p => p.total * p.left_pos / total).sum
-    val rightMean  = pileups.iterator.map(p => p.total * p.right_pos / total).sum
+    val leftMean   = (pileups.iterator.map(p => p.total.toLong * p.left_pos).sum / total).toInt
+    val rightMean  = (pileups.iterator.map(p => p.total.toLong * p.right_pos).sum / total).toInt
     val meanCount  = counts.mean()
 
     new MergedPileup(