feat(mana-ai): OpenTelemetry tracing + Grafana Tempo backend

Add distributed tracing to the mana-ai background runner so mission
execution can be visualized end-to-end in Grafana.

Instrumentation (services/mana-ai/):
- tracing.ts: OTel provider setup with OTLP/HTTP exporter, withSpan() helper
- tick.ts: tick.planMission span with mission/agent/user attributes
- client.ts: planner.complete span with LLM model, tokens, latency

Infrastructure:
- docker/tempo/tempo.yaml: Grafana Tempo config (OTLP HTTP on 4318)
- docker-compose: tempo service + tempo_data volume + mana-ai env var
- docker/grafana/provisioning/datasources/tempo.yml: auto-provisioned

Trace flow:
  tick.planMission (root span)
    └── planner.complete (child span)
        ├── llm.model = "gpt-4o-mini"
        ├── llm.tokens.total = 1234
        └── llm.response.length = 567

Enable: set OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
View: Grafana → Explore → Tempo datasource

Also fixes: removed broken @mana/subscriptions workspace ref from arcade.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

games/arcade/apps/web/package.json

@@ -43,7 +43,6 @@
 		"@mana/shared-icons": "workspace:*",
 		"@mana/shared-stores": "workspace:*",
 		"@mana/shared-tags": "workspace:*",
-		"@mana/subscriptions": "workspace:*",
 		"@mana/shared-tailwind": "workspace:*",
 		"@mana/shared-theme": "workspace:*",
 		"@mana/shared-theme-ui": "workspace:*",