Add GNB benchmark script with dataset-missing sanity check (#579)

[Esma-Ali]

Summary:

• Adds tools/bench_gnb.py to benchmark the repo’s TensorFlow GaussianNBTF implementation.
• When the real EEG dataset isn’t available yet (OpenData S3 “Coming Soon”), the script runs a synthetic sanity-check dataset to validate the model implementation and environment.

Expected behavior right now:

• If no labeled EEG files are found under --data-root, the script prints a dataset-missing note and runs a synthetic sanity check (should typically be >0.90 accuracy).

Why:

• Issue TensorFlow: GaussianNB - Fix Performance #579 reports underperformance and needs a repeatable benchmark to diagnose preprocessing/labeling/model behavior once the dataset is available.

Issue #579