fix fuzz integer overflow in cram encoder.

Input files with very long CIGAR strings and consensus generated
embedded reference can lead to exceptionally long CRAM blocks which
overflow the check for large size fluctuations (to trigger new
compression metric assessments).

Reformulated the expression to avoid scaling up values.

Credit to OSS-Fuzz
Fixes oss-fuzz 68225

cram/cram_io.c

@@ -1984,11 +1984,15 @@ int cram_compress_block2(cram_fd *fd, cram_slice *s,
         // We also get large fluctuations based on genome coordinate for
         // e.g. SA:Z and SC series, but we consider the typical scale of
         // delta between blocks and use this to look for abnormality.
+
+        // Equivalent to (but minus possible integer overflow)
+        //   (b->uncomp_size + 1000)/4 > metrics->input_avg_sz+1000 ||
+        //    b->uncomp_size + 1000    < (metrics->input_avg_sz+1000)/4)
         if (metrics->input_avg_sz &&
-            (b->uncomp_size + 1000 > 4*(metrics->input_avg_sz+1000) ||
-             b->uncomp_size + 1000 < (metrics->input_avg_sz+1000)/4) &&
-            ABS(b->uncomp_size-metrics->input_avg_sz)
-                > 10*metrics->input_avg_delta) {
+            (b->uncomp_size/4 - 750 > metrics->input_avg_sz ||
+             b->uncomp_size         < metrics->input_avg_sz/4 - 750) &&
+            ABS(b->uncomp_size-metrics->input_avg_sz)/10
+                > metrics->input_avg_delta) {
             metrics->next_trial = 0;
         }