Add neural segmentation correctness validation

[raylim]

Adds a correctness validation harness and report for the fixed neural segmentation path now that PR #136 is on main.

Changes:

• Adds a 10-slide neural segmentation correctness panel covering SVS, NDPI, and TIFF slides with varied dimensions, MPP values, and tissue burdens.
• Adds a SLURM wrapper for GPU execution.
• Adds a validation report with acceptance criteria, commands, and observed results.

Validation already run:

• Import-order regression: uv run pytest tests/mussel/test_import_order.py -q -> 3 passed.
• Neural/artifact integration SLURM array 3563392 with qos premium -> 4 selected tasks completed 0:0.
• Tessellate/extract GPU E2E SLURM job 3563398 with qos premium -> 6 passed, 11 deselected.
• New 10-slide panel SLURM job 3563436 with qos premium -> 10/10 passed.

Notes:

• The panel references local filesystem slide paths, so re-running elsewhere requires the same paths or an equivalent manifest.

[raylim]

Closing this because the validation deliverable is the Confluence subpage, not a Mussel repository PR.