New rule: IAM roles trusting AWS accounts not synced by Cartography (external trust)

## Description Add a rule that flags `AWSRole` nodes (in a synced account) whose trust policy allows assumption by an `AWSPrincipal` belonging to an AWS account that Cartography does **not** sync, i.e. an external account. The sync already parses `AssumeRolePolicyDocument`, materializes `(:AWSRole)-[:TRUSTS_AWS_PRINCIPAL]->(:AWSPrincipal)`, and creates stub `AWSAccount` / root-principal nodes for trusted external accounts, so this is a rule-only change with no datamodel or intel work. The reliable marker for "not synced" is `AWSAccount.inscope`: it is set to `true` only for accounts actually in the sync scope; external stub accounts created from trust-policy parsing do not have it. Detection sketch: ```cypher MATCH (a:AWSAccount {inscope: true})-[:RESOURCE]->(role:AWSRole)-[:TRUSTS_AWS_PRINCIPAL]->(p:AWSPrincipal) MATCH (ext:AWSAccount)-[:RESOURCE]->(p) WHERE coalesce(ext.inscope, false) <> true RETURN role.arn, role.name, a.id AS account_id, ext.id AS external_account_id, p.arn AS trusted_principal_arn ``` Output fields: role arn/name, role account id, external (unsynced) account id, trusted principal arn. Consider an allowlist hook (known third-party / partner account ids) as a follow-up; ship the raw detection first. ## Motivation Roles trusting accounts outside the synced perimeter are a silent lateral-movement / persistence vector, often left over from one-off integrations or audits. Keying off `inscope` rather than ARN string parsing makes the detection robust: anything Cartography does not actually sync is treated as external. The graph already encodes the relationship; surfacing it requires no new collection. ## Alternatives Considered - Comparing the account id embedded in the principal ARN against the role's account id: brittle and misses multi-account-in-scope setups. Using `inscope` directly reflects what Cartography actually knows about. - Modeling this as an attack path rather than a flat rule: deferred. A direct rule is the smallest useful increment and composes with attack-path tooling later. ## Relevant Links - `cartography/intel/aws/iam.py` (`transform_role_trust_policies`, `load_external_aws_accounts`) - `cartography/models/aws/iam/role.py` (`TRUSTS_AWS_PRINCIPAL`) - `cartography/models/aws/account.py` (`inscope` property)