Rework diffusion models

[henrikjacobsenfys]

The changes are described here: #190

I'm missing some tests, but the code is more or less how I want it to be, so a review would be nice

[codecov]

Codecov Report

❌ Patch coverage is 95.92326% with 17 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.13%. Comparing base (fff0285) to head (38ecbae).

Files with missing lines	Patch %	Lines
...mple_model/diffusion_model/diffusion_model_base.py	93.20%	5 Missing and 2 partials ⚠️
...mics/sample_model/diffusion_model/delta_lorentz.py	97.59%	3 Missing and 3 partials ⚠️
src/easydynamics/analysis/fit_binding.py	0.00%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #191      +/-   ##
===========================================
- Coverage    98.42%   98.13%   -0.29%     
===========================================
  Files           50       51       +1     
  Lines         3175     3542     +367     
  Branches       576      652      +76     
===========================================
+ Hits          3125     3476     +351     
- Misses          26       36      +10     
- Partials        24       30       +6     

Flag	Coverage Δ
integration	44.91% <35.01%> (-1.26%)	⬇️
unittests	98.02% <94.96%> (-0.41%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[rozyczko]

Thread 1 — Double-counting lorentzian_width in get_all_variables()

Good catch, and well fixed.

Thread 2 — Aggressive reset on Q change in SampleModel._on_Q_change()

I think your reasoning sounds good for now. The key points:

1. Changing Q fundamentally invalidates Q-dependent parameters (per-Q A_0, per-Q lorentzian_width) — there's no reliable way to determine which subset remain valid after the change.
2. The user is explicitly warned when changing Q (it requires confirm=True).
3. The # This may be too aggressive comment serves as a good reminder that this could be revisited.

My suggestion would be to leave the current behaviour as-is for this PR (since it's documented that Q changes are destructive), but perhaps add a note in the docstring of SampleModel.Q setter or _on_Q_change() that clarifies this trade-off explicitly for users. Something along the lines of:

"Changing Q will reset all diffusion model parameters that depend on Q (per-Q amplitudes and widths). This is necessary because the validity of those parameters cannot be determined after a Q change. To avoid accidental data loss, Q changes require confirm=True."

[rozyczko]

lgtm