Shareable experiment URL (encode dataset + models + config)

[jeanclawd]

Problem

There is no way to share or reproduce a specific experiment. "I trained Ridge, SVR-RBF and Gradient Boosting on Airfoil with 10-fold CV and Ridge won" today requires a screenshot or a verbal walkthrough.

Proposal

Encode the experiment state in the URL so a peer can open the link and land on the same setup:

```
https://scikit-learner.app/?dataset=airfoil&target=scaled_sound&features=freq,angle,velocity&models=ridge,svr_rbf,gb&cv=10
```

When the page loads with these params:

1. Auto-load the dataset.
2. Pre-select target and features.
3. Pre-select the model checkboxes.
4. (Optional) Auto-train on load if a `?run=1` flag is present.

Scope

• v1: built-in datasets only (8 keys already enumerated in `load_sample()`).
• v2: custom CSV via content-hash + in-browser localStorage cache (no server). Out of scope here.

Implementation notes

• `URLSearchParams` on page load in `frontend/js/app.js`.
• Add helpers `encodeExperiment(state)` / `decodeExperiment(params)` to keep parsing in one place.
• A "Share" button next to "Export" that copies the current URL via `navigator.clipboard`.

Acceptance criteria

• Opening a shared URL reproduces the exact UI state (dataset, target, features, model selection, CV folds).
• Share button writes a working URL to the clipboard.
• Invalid / unknown params are ignored gracefully (no UI break).

Strategic rationale

Unlocks teaching, blog content, benchmark cases on the landing page, and team collaboration without any backend.