fix: handle wrapped dashboards and inject inputs in portables converter

[kcorbett-hdx]

Summary

• grafana_gen.py: Unwraps {"dashboard": {...}} bundle format before processing dashboards, injects the required datasource entry into __inputs when missing, adds constant template variables from templating.list as __inputs, and always includes the datasource in resources.gfo.yaml inputs
• discoverer.py: Unwraps {"dashboard": {...}} format when reading source dashboards, falls back to extracting constant variables from templating.list as inputs when __inputs is absent

Context

Bundle-format dashboards use a {"dashboard": {...}} wrapper and store table/datasource references as plain constant variables (e.g. raw_table = "__PROJECT_NAME__.__TABLE_NAME__") rather than Grafana __inputs. The portables converter previously produced output dashboards with no __inputs beyond the datasource, and resources.gfo.yaml with no inputs entries — both required for CaC deployment.

Test plan

• Run python3 scripts/bundle_to_yaml.py --source aws/cdn-insights --verbose and confirm all 4 phases pass
• Verify generated dashboard JSONs contain __inputs with datasource + all constant template variables
• Verify resources.gfo.yaml contains an inputs map for each dashboard
• Confirm trafficpeak/security still converts correctly (dashboards already have __inputs — existing path unchanged)

🤖 Generated with Claude Code