feat: infer samples required for built-in lag transforms updates

[jmoralez]

Description

We had to rely on users providing keep_last_n to limit the number of samples kept for each serie for the feature updates because we only supported functions, but since the introduction of the lag_transforms module, we know how many samples are needed to correctly perform the updates, so we can now infer them for the cases where the users exclusively use these transformations, potentially speeding up the predict step by around 3x.

Checklist:

• This PR has a meaningful title and a clear description.
• The tests pass.
• All linting tasks pass.
• The notebooks are clean.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codspeed-hq]

CodSpeed Performance Report

Merging #445 will improve performances by ×3.1

_{Comparing update-samples (7367581) with main (8c28a7f)}

Summary

⚡ 4 improvements
✅ 8 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	update-samples	Change
⚡	test_predict[None-1-False]	729.8 ms	231.8 ms	×3.1
⚡	test_predict[None-1-True]	796.4 ms	286.1 ms	×2.8
⚡	test_predict[None-2-False]	734.4 ms	235.9 ms	×3.1
⚡	test_predict[None-2-True]	801.2 ms	290.1 ms	×2.8