Make StopWindow more efficient and rename it to ConvergenceChecker

[bqth29]

💬 Pull Request Description

When the Stop Window detects that an agent has converged, it is stored in the window's memory and a boolean tensor is sent back to the optimizer to indicate which agents still have not converged. Thus, the oscillators corresponding to converged agents are removed which reduces the computation size and time.

It also becomes more relevant to rename StopWindow as ConvergenceChecker so its purpose is more easily understandable. Documentation has been updated.

Besides, the use_window parameter used in the various optimization functions and methods has been renamed early_stopping for relevance's sake.

✔️ Check list

• The code matches the styling rules
• The new code is covered by relevant tests
• Documentation was added

🚀 New features

None.

🐞 Bug fixes

None.

📣 Supplementary information

None.

⚠️ Breaking changes

Renaming StopWindow and use_window to respectively ConvergenceChecker and early_stopping introduced breaking changes with respect to versions <= 1.2.1.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (a9ab8bb) to head (c7984b6).

Additional details and impacted files

@@            Coverage Diff            @@
##              main       #55   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           36        36           
  Lines         1600      1537   -63     
=========================================
- Hits          1600      1537   -63     

Flag	Coverage Δ
	100.00% <100.00%> (ø)

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.

[bqth29]

Possible bug when combining this new window with the keyboard interrupt feature from #51: if the user stops the optimization manually between the moment when the window is reduced but not the oscillators in the optimizer, this may create shape conflicts when combining both in the final spins that are returned. Need to discuss this.