[Appliance Core] Add safety gating, autonomy modes & action whitelist stub

[saviornt]

Description

Implement autonomy mode enforcement and basic action whitelisting to prevent unsafe or unapproved actions in lower autonomy levels.

Why: Safety-first principle — edge must never take dangerous actions without governance.

Type

• Task

Focus Area (pick one)

• Auth & Security / Appliance Core (Pi edge)

Priority

• Critical

Acceptance Criteria

• AutonomyMode enum (ADVISE, MONITOR, ASSISTED, AUTONOMOUS, etc.)
• SafetyGate class that checks mode + whitelist before any action execution
• Rejected actions logged with clear reason
• Configurable via Pydantic Settings (e.g. current mode from env/config)
• Stub methods for future rollback snapshot integration

Blocker / Dependencies

• Appliance async loop implementation

Notes / Links

• Aligns with security_design.md and vision.md fail-open principles