providers now per-agent: z.ai rides, bifrost stays home

Vanguard gets his own inference pipe. z.ai GLM Coding Plan, /paas/v4.
BifrostClient learned to leave versioned URLs alone. ProviderRegistry
dispatches per agent; every engine asks who it's talking to before
dialing. Annie and Hal none the wiser.

src/api/models.rs

@@ -94,6 +94,22 @@ pub struct SouveraineConfig {
     pub archivist_enabled: bool,
     pub archivist_threshold: f32,
     pub sensorium_bandwidth: String,
+    /// Per-agent model override for the subconscious (N+1) pass. When set,
+    /// it wins over the global `[subconscious] model`. None falls back to
+    /// the global setting, then the agent's own primary model.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub subconscious_model: Option<String>,
+    /// Per-agent model override for the reflection (N+25) pass.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub reflection_model: Option<String>,
+    /// Per-agent model override for the archivist (N+100) synthesis pass.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub archivist_model: Option<String>,
+    /// Per-agent inference provider name (e.g. "zai"). When set, the agent's
+    /// requests route through the named provider instead of the global
+    /// `[inference] provider`. None falls back to the global default.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub provider: Option<String>,
 }
 
 #[derive(Debug, Deserialize)]