Restructure metrics to use tags

[cescoffier]

I was looking at the metrics output and found that the raw HTTP ones are formatted differently:

# HELP http_client_requests_seconds  
# TYPE http_client_requests_seconds summary
http_client_requests_seconds_count{clientName="api.openai.com",method="POST",outcome="SUCCESS",status="200",uri="/chat/completions",} 2.0
http_client_requests_seconds_sum{clientName="api.openai.com",method="POST",outcome="SUCCESS",status="200",uri="/chat/completions",} 4.791650208
# HELP http_client_requests_seconds_max  
# TYPE http_client_requests_seconds_max gauge
http_client_requests_seconds_max{clientName="api.openai.com",method="POST",outcome="SUCCESS",status="200",uri="/chat/completions",} 2.70089925

while we have:

# HELP langchain4j_aiservices_TriageService_triage_seconds
# TYPE langchain4j_aiservices_TriageService_triage_seconds summary
langchain4j_aiservices_TriageService_triage_seconds_count 2.0
langchain4j_aiservices_TriageService_triage_seconds_sum 4.992278791
# HELP langchain4j_aiservices_TriageService_triage_seconds_max
# TYPE langchain4j_aiservices_TriageService_triage_seconds_max gauge
langchain4j_aiservices_TriageService_triage_seconds_max 2.706755083

I think we need to align. In Quarkus, we use tags massively to reduce the number of keys. So, it might be better to follow that principle and have:

langchain4j_aiservices_seconds{aiservice=TriageService, method=triage}

[geoand]

Sure, we can definitely do that

[geoand]

#155 takes care of this