whisper : make beam candidate sort more stable (#1943)

All else being otherwise equal, this encourages the beam candidate
selection to re-use the same decoder, which slightly
reduces the cache size.

I wouldn't expect it to make much of a performance difference,
but it helps when debug printing the cache and beam.

Added as part of understanding #1941.

Author: josharian

whisper.cpp

@@ -5357,7 +5357,10 @@ int whisper_full_with_state(
                             beam_candidates.begin(),
                             beam_candidates.end(),
                             [](const beam_candidate & a, const beam_candidate & b) {
-                        return a.sequence.sum_logprobs_all > b.sequence.sum_logprobs_all;
+                        if (a.sequence.sum_logprobs_all != b.sequence.sum_logprobs_all) {
+                            return a.sequence.sum_logprobs_all > b.sequence.sum_logprobs_all;
+                        }
+                        return a.decoder_idx < b.decoder_idx;
                     });
 
                     uint32_t cur_c = 0;