feat: normalized config schema with PersonaRecord.provider and SPROUT_AGENT_MODEL

[wpfleger96]

Wes's 24 sprout-agent personas all failed with SPROUT_AGENT_PROVIDER required even though Databricks was configured — because the desktop never injected provider/model env vars at spawn time, the goose config fallback only understood the old flat format, and pack import always stored an empty env_vars map. This PR fixes all three structural gaps.

sprout-agent gets a universal SPROUT_AGENT_MODEL env var so a single model value works regardless of which LLM provider is active. GooseDatabricksConfig is extracted into an isolated goose_compat submodule and extended to parse the newer active_provider + nested providers: format Wes's config uses. PersonaRecord gains a first-class provider field alongside runtime and model, completing the normalized config schema, and KnownAcpRuntime metadata is now fully wired into the spawn path.

• SPROUT_AGENT_MODEL added as universal model fallback in Config::from_env() for all three provider arms; KnownAcpRuntime for sprout-agent updated to model_env_var: Some("SPROUT_AGENT_MODEL")
• provider_env_var injection wired into spawn_agent_child() respecting provider_locked; !supports_acp_model_switching guard removed (conflation of bootstrap vs. post-session model switching)
• runtime_env_vars() extended to emit SPROUT_AGENT_MODEL/SPROUT_AGENT_PROVIDER for sprout-agent personas vs. GOOSE_MODEL/GOOSE_PROVIDER for goose; pack import now populates PersonaRecord.env_vars (was always empty) and PersonaRecord.provider
• LLM Provider field added to PersonaDialog for manual persona creation