feat: enable Markdown (.md) file upload in KB Accelerator

Moved md from Phase 2 extensions to allowed formats, added extraction
handler (reuses txt handler), and updated plan_limits defaults to
include md for all plans.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app/api/endpoints/kb_accelerator.py

@@ -57,6 +57,7 @@ MAX_UPLOAD_SIZE = 10 * 1024 * 1024
 
 ALLOWED_EXTENSIONS = {
     "txt": ["text/plain"],
+    "md": ["text/markdown", "text/plain"],
     "docx": ["application/vnd.openxmlformats-officedocument.wordprocessingml.document"],
 }
 
@@ -64,7 +65,6 @@ ALLOWED_EXTENSIONS = {
 PHASE2_EXTENSIONS = {
     "pdf": ["application/pdf"],
     "html": ["text/html"],
-    "md": ["text/markdown", "text/plain"],
 }
 
 
@@ -110,7 +110,7 @@ async def _check_lifetime_limit(user: User, limits: PlanLimits, db: AsyncSession
 
 
 async def _check_format_allowed(source_format: str, limits: PlanLimits) -> None:
-    allowed = limits.kb_allowed_formats or ["txt", "paste"]
+    allowed = limits.kb_allowed_formats or ["txt", "paste", "md"]
     if source_format not in allowed:
         raise HTTPException(
             status_code=status.HTTP_403_FORBIDDEN,
@@ -205,7 +205,7 @@ async def get_quota(
             kb_accelerator_enabled=False,
             lifetime_conversions_used=committed_count,
             lifetime_conversions_limit=0,
-            allowed_formats=["txt", "paste"],
+            allowed_formats=["txt", "paste", "md"],
             detailed_analysis=False,
             conversational_refinement=False,
             step_library_matching=False,