feat(platform): course-plan student visibility, multi-voice TTS, OCR sources, branches

Backend (encoach_ai_course):
- workbook_attempt model + scoring + REST endpoints for student attempts
- dialogue_parser splits scripts by speaker, classifies gender, strips labels
- media_service: multi-voice TTS via Polly/ElevenLabs, ffmpeg concatenation,
  manual media upload endpoint (audio/image/video) with size validation
- source_indexer: OCR fallback (pytesseract + pdf2image) for scanned PDFs,
  page-streaming to stay under memory limit
- exercise_extractor + rag_context for RAG-grounded interactive workbooks
- course_plan_pipeline: v2 generator that grounds week material on indexed
  sources and persists grounded_on_json metadata
- security: access rules for new models

Backend (encoach_lms_api):
- branches model + controller (entity-scoped LMS branches)
- classroom_ext + course_ext (assignment + section workflow)
- classrooms controller: students/teachers/assign-course endpoints

Frontend:
- StudentDashboard: surface assigned AI course plans alongside enrollments;
  enrolled-courses stat now counts plans+enrollments
- InteractiveWorkbook + PlanReader components
- AdminCoursePlanDetail: media drawer with upload buttons (audio/image/video),
  hidden file inputs, upload mutation
- AdminBranches page + sidebar entry
- coursePlan/lms/classrooms services + types updated for new endpoints
- i18n: studentDash.myCoursePlans/noCoursePlans (en + ar)

Infra & docs:
- odoo.conf: bump memory limits to 4G/5G for OCR + sentence-transformers
- .gitignore: ignore *.tsbuildinfo
- docs/ASSIGNMENT_WORKFLOW.{md,pdf}
- smoke_*.py end-to-end tests for assignment workflow, entity isolation,
  course-plan RAG pipeline

Made-with: Cursor

odoo.conf

@@ -19,8 +19,12 @@ limit_time_real = 900
 ; "submit exam with all generated questions" (≈2-5 MB JSON) no longer fail
 ; with HTTP 413. Default is 4 MB; 128 MB covers videos + bulk question JSON.
 limit_request = 134217728
-limit_memory_hard = 2684354560
-limit_memory_soft = 2147483648
+; OCR (tesseract + 200dpi rasterization) + sentence-transformers (~80 MB
+; model + torch tensors) + Odoo baseline routinely cross 2 GB even with
+; per-page streaming. Bump soft to 4 GB / hard to 5 GB so a 100-page
+; scanned workbook doesn't trigger a worker reload mid-index.
+limit_memory_hard = 5368709120
+limit_memory_soft = 4294967296
 db_maxconn = 32
 data_dir = /tmp/odoo-data