Add typed config loader and validation for safe defaults

[Aggredicus]

Add typed config loader and validation for safe defaults

Purpose

Create a typed configuration layer so config.example.yaml and runtime behavior stay aligned and dangerous settings are validated before use.

The project already documents safe defaults such as summon-only public behavior and no automatic multi-packet replies. This issue turns those expectations into validated config behavior.

Scope

Add or design:

src/permanet_agent/config.py

Use Pydantic or another existing dependency already in the project to validate config.

Validate fields such as:

• summon phrases
• radio adapter
• public channel mode
• max response chars
• cooldown seconds
• allow automatic multipacket replies
• AI backend
• storage path

Safety requirements

Defaults must remain safe:

radio.adapter = mock
policy.public_requires_summon = true
policy.allow_automatic_multipacket_replies = false

Dangerous or unsupported values should fail clearly.

Non-goals

Do not add live radio behavior.
Do not add private group routing.
Do not implement full Issue #1 response behavior unless needed for config tests.

Acceptance criteria

• Typed config models exist.
• Config can be loaded from YAML.
• config.example.yaml validates successfully.
• Unsafe values fail with clear errors.
• Tests cover valid config and unsafe config.
• Docs explain config validation and safe defaults.

Rollback plan

Keep config loading isolated so it can be reverted without changing routing behavior.