Fix for Divide-by-zero in ixheaacd_lpc_to_td

This changes handle the Divide-by-zero runtime error
reported while calculating the gain because
imaginary and real data were zero.

Bug: ossFuzz:61788
Test: poc in bug
diff --git a/decoder/ixheaacd_tcx_fwd_mdct.c b/decoder/ixheaacd_tcx_fwd_mdct.c
index 267af0d..3fd55e8 100644
--- a/decoder/ixheaacd_tcx_fwd_mdct.c
+++ b/decoder/ixheaacd_tcx_fwd_mdct.c
@@ -165,7 +165,7 @@
 
   for (i = 0; i < size_n / 2; i++) {
     gains[i] =
-        (FLOAT32)(1.0f / sqrt(data_r[i] * data_r[i] + data_i[i] * data_i[i]));
+        (FLOAT32)(1.0f / (sqrt(data_r[i] * data_r[i] + data_i[i] * data_i[i]) + FLT_EPSILON));
   }
 
   return;