fixing issue in Viterbi when decoding zero codeword

it seems that when the input is only zero, the calculated max
becomes zero too. This causes a div by zero later in the
normalization (q->gain_quant/max).

lib/src/phy/fec/viterbi.c

@@ -528,8 +528,7 @@ int srslte_viterbi_decode_f(srslte_viterbi_t *q, float *symbols, uint8_t *data,
     len = 3 * (frame_length + q->K - 1);
   }
   if (!q->decode_f) {
-    
-    float max = -9e9; 
+    float max = 1e-9;
     for (int i=0;i<len;i++) {
       if (fabs(symbols[i]) > max) {
         max = fabs(symbols[i]);