Add binary decoding and correct axis name for plot, fix plot per layers

[anandawolz]

Updates after @CeliaBenquet comments on this PR #59:

1. decoding() signature: changed to return a DecodeResult namedtuple (overall_score, per_label_error, per_label_score, label_types) instead of a raw 3‑tuple

2. Compute method storage: Updated Decoding.compute() to collect and store full DecodeResult objects for each layer, rather than unpacking into separate lists. Callers now work with a consistent, typed return value.

3. DecodingPlot:

• In plot(), derive the base title (“Layer‑wise decoding performance” or “Layer‑wise decoding error”) and the y‑axis label from the new DecodeResult.label_types and the plot_error flag—no more deprecated flags or special cases.
• _transform() was adapted to the change in 1 & 2
  Comment: The custom title specified by the user is still not applied when calling plot_metric. I haven’t investigated this issue yet.

4. Test updates:

• test_decoding_function now expects a DecodeResult and unpacks its fields.
• test_decoder_plot adds a tiny _LayerMap stub implementing .tolist() so that the single‑layer vs. multi‑layer plot branches still call plot_decoding vs. plot_layer_decoding correctly.
• no new tests added, but tests are passing

5. Requirements: Please don't merge it: works in one container, breaks in another.

!! I still have to standardize the DecodeResult when the decoding function for dataset_label == visual or == HPC is called. For now their respective functions: decoding_frames& decoding_pos_dir are still returning tuples.

Original PR comments:

This PR refactors and extends the decoding and plotting utilities to properly handle both regression and classification (binary/multiclass) targets in neural decoding analysis, as well as adds options to the plotting interface for binary outcomes. It also makes sure that the test passes.

Details:

1. Decoding for specific label type. Decoding() now auto-detects label type (regression/classification) and applies appropriate metrics:

• Regression: R² and median error
• Classification: Accuracy and classification error
• Decoding score is R² (continuous) or accuracy (binary)

2. Plot.

• Adapted the y-axis name & title for binary decoding r^2-score --> accuracy
• Improved plotting: Decoding.plot() now dispatches to the correct helper (plot_decoding vs. plot_layer_decoding) based on the result shape

3. Readability. Replace manual k-loop and 8/9-split with a single GridSearchCV for better readability, runs a bit slower now

4. Test suite fixes.

• Enhanced _cut_array and activation post‑processing to handle both 2D and 3D activation arrays without indexing errors
• Added get_params/set_params stubs to the DummyKNNDecoder mock so it can be cloned by GridSearchCV

[CeliaBenquet]

nice! thanks @anandawolz, overall, some formating and comments to clarify + the labels_is_binary tag to remove as discussed, to be replaced with dict in the metrics directly.

[CeliaBenquet]

@anandawolz can you resolve the last comments and confirm that the requirements.txt works in all containers now? then I will merge, thank you!

[anandawolz]

Hi @CeliaBenquet, I’ve rerun the full pipeline on the hippocampus dataset and everything passes.
requirements.txt has been reverted to your original version - my accidental changes are removed and I haven’t modified anything else. From my side, this PR is ready to be merged.

[CeliaBenquet]

@anandawolz can you resolve the review comments as well :)

[anandawolz]

@CeliaBenquet, done resolving the review comments