fix(models): MLPRegression reads its hyperparameters from self.params, so HPO never reaches training

[Maarmapa]

MLPRegression kept its configuration in two places: self.params, the dict of construction kwargs, which is what train, save and load read; and instance attributes, which is what ModelFactory._process_param and both optimizers write via setattr(obj, key, value).

A search over hidden_size, activation, learning_rate or epochs therefore never reached the network: every trial rebuilt the construction-time model, the study reported a best trial, and the model it produced had never been trained with the values that won. Nothing failed while that was true.

Mirror the configuration onto instance attributes in __init__ and read those everywhere. Two smaller defects fall out of the same split and are fixed here too:

• hidden_size fell back to 100 in train and to 5 in load, against a schema that declares 16. _CONFIG_DEFAULTS is now the single home for every fallback, and a test pins it to the schema field by field.
• save persisted the construction kwargs next to post-search weights, so reloading rebuilt a network of the wrong width. It now records the configuration the model actually ran with.

Also fixes an AttributeError that made the model unusable outside ModelFactory: train reads self.log_train_every_n_steps and its three siblings, which __init__ never set.

Adds tests/back/models/test_mlp_regression.py, including one that drives a real OptunaOptimizer over hidden_size and asserts the widths of the networks actually built equal the widths Optuna suggested.

Closes #842

Claude-Session: https://claude.ai/code/session_01BzZ6td4VJdRDe76UndWZY3

Summary

Short explanation of what changed and why.

---

Type of Change

Check all that apply like this [x]:

• Backend change
• Frontend change
• CI / Workflow change
• Build / Packaging change
• Bug fix
• Documentation

---

Changes (by file)

Briefly list the important modified files and what was done.

Example:

• path/to/file.py: description
• another/file.tsx: description

---

Testing (optional)

Only add if there's something reviewers should verify.

---

Notes (optional)

Additional context or considerations.