feat(server): llama-server health probe

Second HealthProbe (after disk-watermark): liveness via GET /health on the
configured llama base URL + tokens/sec degradation watch derived from recorded
InferenceCompletedEvents (windowed average vs configurable threshold, opt-in via
llamaTpsWarnBelow). Plugs into the existing HealthMonitor edge-detection loop;
no new events. Observability spec §4.

core/config/src/main/kotlin/com/correx/core/config/CorrexConfig.kt

@@ -24,6 +24,11 @@ data class CorrexConfig(
  * the on-disk footprint every [intervalMs] and records a degraded/restored event on the edge.
  * [diskWarnBytes] is the absolute event-log + CAS size that trips a warning; [diskWarnGrowthBytesPerMin]
  * is the growth rate that warns log compaction is becoming non-theoretical. Defaults suit a homelab.
+ *
+ * [llamaLivenessTimeoutMs] bounds the GET /health ping to the llama-server; a connection error or
+ * timeout within this window is reported as DEGRADED (liveness). [llamaTpsWarnBelow] is the
+ * tokens/sec floor below which the probe reports degradation (KV-cache pressure, thermal throttle);
+ * 0.0 disables the degradation dimension (no baseline to compare against on a fresh install).
  */
 @Serializable
 data class HealthConfig(
@@ -31,6 +36,8 @@ data class HealthConfig(
     val intervalMs: Long = 30_000,
     val diskWarnBytes: Long = 5_000_000_000,
     val diskWarnGrowthBytesPerMin: Long = 50_000_000,
+    val llamaLivenessTimeoutMs: Long = 5_000,
+    val llamaTpsWarnBelow: Double = 0.0,
 )
 
 /**