fix(agents): use warehouse-backed query surfaces

[jonathanhaaswriter]

Summary

• switch agent query/get-asset tools and their CLI/job wiring from Snowflake clients to the generic warehouse read surface
• add a lightweight app.OpenWarehouse helper so direct cerebro query uses the configured warehouse backend without booting the full app
• add regression coverage for warehouse-backed agent queries and direct sqlite-backed CLI queries

Validation

• go test ./...
• GOTOOLCHAIN=go1.26.1 go run github.com/golangci/golangci-lint/v2/cmd/golangci-lint@v2.8.0 run --timeout 5m ./...
• go run ./scripts/check_agent_sdk_contract_compat/main.go --require-baseline --base-ref=writer/main
• make agent-sdk-docs-check
• make agent-sdk-packages-check
• go run ./scripts/check_api_contract_compat/main.go --require-baseline --base-ref=writer/main
• make api-contract-docs-check
• make config-docs-check
• make devex-codegen-check
• go test ./internal/app -run "Test(PreCommitHookRunsFastLintOnStagedGoFiles|PrePushHookRunsChangedDevexPreflight|DockerBuildCommandsPassGoVersionBuildArg|DevelopmentGuideDocumentsDevexPreflight|DevexScriptPlansRelevantChecks|DevexScriptChangedModeIncludesWorkspaceDiffSources)"
• make openapi-check
• go run ./scripts/check_report_contract_compat/main.go --require-baseline --base-ref=writer/main
• make report-contract-docs-check
• make vendor-check

[jonathanhaaswriter]

I think get_asset_context regressed in two ways:

• the warehouse GetAssetByID path is querying id, not the canonical _cq_id that the old Snowflake implementation used, so valid lookups can now miss rows
• a miss now comes back as a marshalled null payload instead of a not-found error, because the warehouse implementations return (nil, nil)

Can we keep _cq_id semantics here and normalize misses back into the previous not-found behavior?