feat(gradebook): gradebook overhaul - weights, external assessments, CSV import

[LWS49]

Summary

This PR ships the complete gradebook overhaul: a new gradebook page with column picker and CSV export; a weighted view with configurable per-tab weights, keep-highest rules, and projected totals; an externally-graded assessments system (add/rename/delete columns, per-student inline grade editing); and a CSV bulk import wizard that lets professors define component metadata, download a template, upload filled marks, verify a preview, resolve grade conflicts, and commit in one flow.

Externals are surfaced as a synthetic "External Assessments" category in the gradebook API response: the backend assembles them from course_external_assessments and serializes them with negative IDs (distinguishing from real assessment IDs), avoiding any schema change to the existing categories/tabs structure. Weight contributions are stored in a new course_gradebook_contributions table that accepts both real assessment_id entries and an external_assessment_id slot; the weighted total is a canvas-style additive sum (no normalization cap, bonus allowed).

The CSV import uses a two-gate flow: preview (dry-run — resolves every identifier against the current roster, returns a sample + conflict list, writes nothing) followed by commit (re-validates inside one transaction, authoritative). Re-importing into an existing component upserts grades only — max marks and weightage are never touched. The binding key is the resolved course_user_id; the raw identifier string is snapshotted into imported_identifier for audit and reassigned-ID mismatch detection.

Design decisions

• Synthetic negative-ID serialization for externals - The gradebook index view assembles external assessments into a virtual category with negative IDs rather than polluting the real assessment tree. This lets the frontend treat externals uniformly in the column picker, weighted table, and CSV export without a schema migration on existing categories/tabs.

• Component metadata in the dialog, not the CSV - The import wizard collects name, weightage, and max marks in a dialog form; the CSV carries only Identifier + one grade column per component. This avoids professors having to format a structured header row with embedded metadata and keeps the CSV human-editable with a plain template.

• preview + commit two-gate instead of a single upload - commit re-validates entirely server-side in one transaction and never trusts a client-sent conflict list. The client posts the same payload both times; the preview result is purely advisory. This prevents a race between preview and commit (roster changes, concurrent imports) from silently committing bad data.

• Conflict UI is a design copy, not a shared component - ExternalGradeConflictPrompt/Table mirror the visual design of ExternalIdConflictPrompt/Table (user-invitations) but are independent components acting on grades. Sharing the component would couple two unrelated domains through a generic prop interface neither actually owns.

Regression prevention

Covers: gradebook load and column picker toggle, CSV export output shape, weighted total calculation (equal/custom/keep-highest), external assessment CRUD, per-student inline grade edit, import service (preview dry-run, fresh import, upsert keep/replace, blank cell handling, conflict detection, identifier mismatch, determinacy), import controller (authorization, malformed grades, duplicate components, email mode, conflicts in response), template CSV escaping, wizard step flow (no conflicts, unresolved error, conflict prompt keep + replace paths).

Manual testing confirmed: all 8 scenarios ran including both conflict resolution paths (Keep Existing + Replace), email-mode import, and the error path with an unresolved student ID.

Existing gradebook controller, external assessments controller, and contribution model specs updated to cover the new behaviour. No changes to existing assessment or submission write paths.