rank: selected_and_all_and_picture crashes when Matrix metric is missing

[avinxshKD]

What happened:

selected_and_all_and_picture assumes every test result has a Matrix metric.

For normal benchmarks with scalar metrics only, _draw_pictures() does this:

python
matrix = test_result.get("Matrix")
for key in matrix.keys():

So matrix is None and rank saving crashes after the benchmark result is already available.

What you expected to happen:

Rank CSV files should still be saved.
If picture mode needs Matrix, Ianvs should either skip picture generation for results without Matrix, or fail with a clear error.

How to reproduce it:

Use save_mode: selected_and_all_and_picture with a result that only has scalar metrics, for example:

{"accuracy": 0.9}

Actual error:
AttributeError: 'NoneType' object has no attribute 'keys'

Anything else we need to know?:
This is in core/storymanager/rank/rank.py, inside _draw_pictures().

[avinxshKD]

/assign