Feat/room blockout

[ekremarmagankarakas]

Summary

• Fix merge group scheduling — replaced high-weight edge approach with node contraction before DSATUR coloring, guaranteeing merged courses always share the same time slot
• Add room blockout feature — users can upload an optional CSV specifying (room, day, block) slots where rooms are unavailable; the scheduler respects these during room assignment
• Surface unroomed exams — courses that have a time slot but no available room (all rooms blocked) are now tracked, persisted to the DB, and visualized distinctly from unscheduled courses

Room blockout CSV support:

• RoomBlockoutSchema and RoomBlockoutAdapter with flexible day/block
  parsing (accepts integers or named values like 'Monday', '9AM-11AM')
• room_blockouts field on SchedulingDataset; DatasetFactory accepts
  optional blockouts_df
• _assign_rooms skips blocked rooms and tracks unroomed courses in
  ScheduleResult.unassigned so unplaced_exams count is accurate
• API route and DatasetService accept optional room_blockouts file
• Precompute blockout_slots into file_paths JSONB metadata at upload
  time so schedule retrieval never re-reads from S3
• Frontend: optional upload slot with badge, store, and API support
• Tests: adapter parsing, scheduler enforcement, end-to-end pipeline

Merge group scheduling fix:

• Replace weight-9999 edge hack with DSATUR graph contraction — edges
  mean must-be-different-color so the old approach prevented merged
  courses from sharing a slot rather than guaranteeing it
• Contract each merge group into a virtual node, union external edges,
  expand colors back after coloring
• Sum soft-constraint penalties across all CRNs in a group during slot
  selection so back-to-back and load costs reflect the full group
• Add algorithm correctness test suite (coloring validity, no hard
  conflicts, merge invariants, determinism)

Frontend:

• Pass per-slot blocked room counts through schedule API response
  (blockouts field on both generate and retrieve endpoints)
• CompactView shows blocked room badge per calendar cell
• DensityView shows blocked room badge alongside conflict badges
• StatisticsView shows blockout stats card (rooms constrained, total
  blocked slots) when blockouts are present
• Extract extractTimeFromBlock to shared utils.ts, removing duplicate
  implementations in both calendar views

Other:

• Run scheduler in asyncio.to_thread (CPU-bound work was blocking the
  event loop)
• Make instructor, department, and term optional in course schema so
  datasets missing these fields upload successfully

[MarkFontenot]

Pushed a standalone fix commit (fb4ce4b) for a pre-existing crash + count divergence that this feature's spec depends on — EXENG-14.

POST /api/schedule/generate/{id} 500'd on any over-capacity merge group: _build_generation_response received merges=None (caller omitted it) and called merges.get() in the unscheduled-merge loop. Separately, unplaced_exams/num_classes diverged between generate (counted only result.unassigned) and retrieve (_calculate_summary_stats, which also counts null-slot unscheduled-merge rows).

• Pass merges into _build_generation_response (+ defensive merges = merges or {}).
• New _summarize_placement() computes counts matching the row-based _calculate_summary_stats.
• Regression tests: backend/tests/services/schedule/test_schedule_summary.py (incl. the generate==retrieve invariant).

Verified live: generate now 200; generate and retrieve both report unplaced_exams=2, num_classes=5 on the repro (was 500 / would've been 0 vs 2).

Pre-existing on develop; fixed here because goal #3 (accurate unplaced_exams, surfacing unscheduled exams) can't ship without it. num_students parity is a separate follow-up (EXENG-15).

[MarkFontenot]

@coderabbitai review

[linear]

EXENG-14