feat(context): reasoning-thread continuity, L3 doc filtering, and gate hardening

Threads reasoning_content across inference calls and journal renders, filters
markdown noise out of L3 repo-knowledge retrieval, adds PlanLinter H3 checks,
and tightens filesystem tool output/dir-listing behavior surfaced by prior
live QA (see project_readloop_campaign memory).

core/inference/src/main/kotlin/com/correx/core/inference/PromptRenderer.kt

@@ -10,6 +10,14 @@ import kotlinx.serialization.Serializable
 data class ChatMessage(
     val role: String,
     val content: String,
+    // Carries ContextEntry.reasoning forward for assistant tool-call turns — see that field's
+    // doc for why this exists (Gemma-family looping when its own prior reasoning is missing).
+    val reasoning: String? = null,
+    // Optional tool-call-id that links a role:tool response to the assistant's tool_calls entry
+    // (sourceId in the event store). Required by the Gemma4 Jinja template to render tool
+    // response blocks with the correct function name; without it every tool response is
+    // rendered as "unknown" and the thought/reasoning channel logic can break.
+    val toolCallId: String? = null,
 )
 
 object PromptRenderer {
@@ -61,5 +69,7 @@ object PromptRenderer {
             EntryRole.USER -> "user"
         },
         content = content,
+        reasoning = reasoning,
+        toolCallId = if (role == EntryRole.TOOL) sourceId else null,
     )
 }