Fix forward phosphosite ODE panel

[bibymaths]

Motivation

• The forward-simulation phosphosite panel was plotting inflated FC-like values instead of the raw ODE phosphosite state, so the panel must read directly from the ODE state vector rather than using pred_fc/simulate_and_measure()/pipeline exports.
• The forward simulations themselves should continue to be generated via simulate_until_steady() for both WT and KO.

Description

• Added extract_phosphosite_states_from_Y to read phosphosite state trajectories directly from the ODE state arrays Y_wt/Y_ko using idx.offset_y, idx.n_sites, and idx.sites, with optional safe t0 normalization.
• Added _safe_min_max helper used by a small debug expander to report raw WT/KO and plotted WT/KO min/max values for the selected protein.
• Plumbed a dedicated checkbox Normalize forward phosphosite states to t0 that controls only the 4th panel and sets the y-axis label to either Raw phosphosite ODE state or Phosphosite ODE state normalized to t0.
• Reworked the 4th (phosphosite state) panel to plot the values from extract_phosphosite_states_from_Y and explicitly avoid using wt_pho, ko_pho, pred_phospho_picked.csv, pred_fc, or simulate_and_measure() for this panel, while keeping the first three forward panels (mRNA, protein, signaling S) and the Data Fit Inspector unchanged.
• Only scripts/compare_mechanisms.py was modified; no networkmodel/ modules were changed.

Testing

• Ran python -m py_compile scripts/compare_mechanisms.py which succeeded.
• Ran git diff --check / basic repository checks which showed no whitespace or diff errors.

---

Codex Task