Expand gathered telemetry: golden signals, traces, business metrics, and service logging #159

Summary

The observability stack (Grafana, Loki, Prometheus, Jaeger, OTel collector) is in place, but the API currently only emits default ASP.NET Core/HttpClient/HotChocolate traces, one custom metric pair (tournaments_created, active_tournaments), and unstructured request logs. This issue tracks expanding what we actually gather so the stack is useful for both API health and tournament-flow business questions.

Golden signals (metrics)

• graphql.server.request.duration histogram, tagged graphql.operation.type + graphql.operation.name — p90/p95/p99 and request rate per GraphQL operation (the ASP.NET Core histogram can't distinguish operations since everything hits /graphql)
• graphql.server.errors_total counter, tagged graphql.operation.name + graphql.error_code — GraphQL errors return HTTP 200, so they're invisible to HTTP-status-based error rate today

Traces

• Add OpenTelemetry.Instrumentation.EntityFrameworkCore (AddEntityFrameworkCoreInstrumentation()) — currently missing, so DB time is invisible inside request traces
• Span for TournamentAutoCloseJob (Hangfire job execution is currently untraced)
• Span around BracketMutations.GenerateBracket to separate algorithm time from persistence time
• Span/event on the optimistic-concurrency retry paths (JoinTournament, AddParticipant, CorrectMatchResult, UpdateRound to see retry counts, not just final outcome

Business metrics

• Tournaments: tournaments_closed_total{reason="manual|auto|full"}, tournaments_deleted_total
• Participants: tournament_join_attempts_total{result="success|full|closed|already_joined"}, slot-contention counter
• Matches/Brackets: matches_played_total, matches_needing_replay_total, match_result_corrections_total, bracket_generation_total{result="success|failure"}
• Auth: login_attempts_total{result="success|failure"}, user_registrations_total, refresh_token_failures_total{reason}

Logging

• Push domain IDs (TournamentId, MatchId, BracketId, etc.) into the Serilog log context via LogContext.PushProperty in BracketMutations, ParticipantMutations, MatchMutations, TournamentMutations, UserMutations, so existing request/error logs become filterable by entity
• Targeted INFO logs (same across environments — verbosity controlled via MinimumLevel config, not env-conditional code) for state transitions that don't surface as a GraphQL error and wouldn't be obvious from a trace span alone:
  • BracketCompletionService — bracket marked complete
  • BracketService — bracket generation outcome (seed/round counts)
  • MatchCascadePositionCalculator — cascade applied, downstream matches affected
  • MatchCorrectionService — correction applied (previous vs. new result), retry count on concurrency conflict

Non-Goals

• Rate-limiter rejection metrics
• Per-field/DataLoader-level custom metrics beyond what HotChocolate's own instrumentation already traces