Connected apps can still read the user's name from SOUL.md with memory off #552

Found while testing #548. That PR keeps the configured name out of memory_remember's answer. But a home named before first start or through onboarding keeps the name in SOUL.md, and that goes into the prompt, so a /v1 turn with openai_api.allow_memory off can still reveal it. The same may apply to incognito chats with memory off.

To look at: what Agent.memory_allowed false leaves in the prompt from SOUL.md and the personality files. Keep Sentient's own personality, and drop or replace anything that identifies the user.

Done when: in a real run on a home named through onboarding, "what's my name?" through /v1 with memory off doesn't reveal it, and normal chats still use the name.