[V1][Spec Decoding] Fix duplicate metrics error

```
ValueError: Duplicated timeseries in CollectorRegistry:
{'vllm:spec_decode_num_accepted_tokens_per_pos_total',
'vllm:spec_decode_num_accepted_tokens_per_pos_created',
'vllm:spec_decode_num_accepted_tokens_per_pos'}
```

Signed-off-by: Mark McLoughlin <[email protected]>

Author: markmc

vllm/v1/spec_decode/metrics.py

@@ -140,15 +140,16 @@ def __init__(self, speculative_config: Optional[SpeculativeConfig],
         num_spec_tokens = (speculative_config.num_speculative_tokens
                            if self.spec_decoding_enabled else 0)
         pos_labelnames = labelnames + ["position"]
+        base_counter = prometheus_client.Counter(
+            name="vllm:spec_decode_num_accepted_tokens_per_pos",
+            documentation="Accepted tokens per draft position.",
+            labelnames=pos_labelnames)
         self.counter_spec_decode_num_accepted_tokens_per_pos: \
             list[prometheus_client.Counter] = []
         for pos in range(num_spec_tokens):
             pos_labelvalues = labelvalues + [str(pos)]
             self.counter_spec_decode_num_accepted_tokens_per_pos.append(
-                prometheus_client.Counter(
-                    name="vllm:spec_decode_num_accepted_tokens_per_pos",
-                    documentation="Accepted tokens per draft position.",
-                    labelnames=pos_labelnames).labels(*pos_labelvalues))
+                base_counter.labels(*pos_labelvalues))
 
     def observe(self, spec_decoding_stats: SpecDecodingStats):
         if not self.spec_decoding_enabled: