agents: multi-step agent support

Split agent prompts on === PROMPT === delimiter. Each step runs as
a new user message in the same LLM conversation, so context carries
forward naturally between steps. Single-step agents are unchanged.

- AgentDef.prompt -> AgentDef.prompts: Vec<String>
- AgentBatch.prompt -> AgentBatch.prompts: Vec<String>
- API layer injects next prompt after each text response
- {{conversation:N}} parameterized byte budget for conversation context

Co-Authored-By: Kent Overstreet <kent.overstreet@linux.dev>

src/subconscious/llm.rs

@@ -21,16 +21,17 @@ pub(crate) fn call_simple(caller: &str, prompt: &str) -> Result<String, String>
         }
     };
 
-    super::api::call_api_with_tools_sync(caller, prompt, None, &log)
+    let prompts = vec![prompt.to_string()];
+    super::api::call_api_with_tools_sync(caller, &prompts, None, &log)
 }
 
-/// Call a model using an agent definition's configuration.
-pub(crate) fn call_for_def(
+/// Call a model using an agent definition's configuration (multi-step).
+pub(crate) fn call_for_def_multi(
     def: &super::defs::AgentDef,
-    prompt: &str,
+    prompts: &[String],
     log: &(dyn Fn(&str) + Sync),
 ) -> Result<String, String> {
-    super::api::call_api_with_tools_sync(&def.agent, prompt, def.temperature, log)
+    super::api::call_api_with_tools_sync(&def.agent, prompts, def.temperature, log)
 }
 
 /// Parse a JSON response, handling markdown fences.