Add a proctored exam grade fact (tfact_exam_grade) to the dimensional layer #2741

Epic: #2073 · Blocks: #2088

Description/Context

Proctored exam grades have no home in the dimensional layer. The only place they surface is marts__micromasters_dedp_exam_grades, which still reads three intermediate models. #2319 and #2403 move semester and passing_grade onto dim_course_run, but #2088 can't close until the grades themselves come from a dimensional fact.

tfact_grade isn't a good fit on its own. The exam grades come from three places, and only one of them is in tfact_grade today:

The current logic for the two MITxOnline paths is in int__mitxonline__proctored_exam_grades, and for MicroMasters in int__micromasters__dedp_proctored_exam_grades. The mart dedups learners who appear in both MicroMasters and MITxOnline for the same exam run; the fact needs to keep that rule.

This was flagged in the July warehouse audit (#2375) and listed in dimensional_warehouse_todo.md, but never written up as an issue.

Plan/Design

1. Check that every exam run with grades has a current dim_course_run row. The fact needs a courserun_fk for each; MicroMasters exam runs that aren't MITxOnline course runs may be missing. If they are, add them to dim_course_run first.
2. Add tfact_exam_grade: one row per learner per exam attempt, with user_fk, courserun_fk, platform_fk, the grade, the source it came from, and the grade timestamp. semester and passing_grade stay on dim_course_run.
3. Leave tfact_grade as is. Adding exam rows there would double-count the dedicated exam runs it already holds.
4. Rebuild marts__micromasters_dedp_exam_grades on tfact_exam_grade, dim_course_run, dim_course and dim_user (which already carries the MicroMasters and MITxOnline emails and usernames), diff it against the current output, and remove it from dimensional_layering_baseline.txt. That closes Migrate marts__micromasters_dedp_exam_grades to use dimensional layer #2088.