memory: add temperature support to agent defs, update reflect prompt

Thread temperature parameter from agent def header through the API
call chain. Agents can now specify {"temperature": 1.2} in their
JSON header to override the default 0.6.

Also includes Kent's reflect agent prompt iterations.

poc-agent/src/api/mod.rs

@@ -70,12 +70,24 @@ impl ApiClient {
         ui_tx: &UiSender,
         target: StreamTarget,
         reasoning_effort: &str,
+    ) -> Result<(Message, Option<Usage>)> {
+        self.chat_completion_stream_temp(messages, tools, ui_tx, target, reasoning_effort, None).await
+    }
+
+    pub async fn chat_completion_stream_temp(
+        &self,
+        messages: &[Message],
+        tools: Option<&[ToolDef]>,
+        ui_tx: &UiSender,
+        target: StreamTarget,
+        reasoning_effort: &str,
+        temperature: Option<f32>,
     ) -> Result<(Message, Option<Usage>)> {
         match &self.backend {
             Backend::OpenAi { base_url } => {
                 openai::stream(
                     &self.client, base_url, &self.api_key, &self.model,
-                    messages, tools, ui_tx, target, reasoning_effort,
+                    messages, tools, ui_tx, target, reasoning_effort, temperature,
                 ).await
             }
             Backend::Anthropic => {

poc-agent/src/api/openai.rs

@@ -21,6 +21,7 @@ pub async fn stream(
     ui_tx: &UiSender,
     target: StreamTarget,
     reasoning_effort: &str,
+    temperature: Option<f32>,
 ) -> Result<(Message, Option<Usage>)> {
     let request = ChatRequest {
         model: model.to_string(),
@@ -28,7 +29,7 @@ pub async fn stream(
         tool_choice: tools.map(|_| "auto".to_string()),
         tools: tools.map(|t| t.to_vec()),
         max_tokens: Some(16384),
-        temperature: Some(0.6),
+        temperature: Some(temperature.unwrap_or(0.6)),
         stream: Some(true),
         reasoning: if reasoning_effort != "none" && reasoning_effort != "default" {
             Some(ReasoningConfig {