feat(sf): SequenceFeature.scale_composition — scale-based baseline featurizer (#307)

[breimanntools]

Adds SequenceFeature.scale_composition — a no-positional-split baseline featurizer. For each sequence it averages every scale over the residues of a span, giving the (n_seq, n_scales) matrix X: the sequence's mean profile in scale-space (the scale-based analogue of amino-acid composition), with no positional information.

Application. Build a baseline feature set for a prediction model and compare the same classifier on this X versus a CPP feature_matrix, to show how much the positional Part-Split-Scale features add over a plain scale average (the "scale baseline vs CPP" comparison). It is not a positional feature set.

Notes:

• Named scale_composition (not scale_mean) to fit the compositional-descriptor family — see the follow-up issue Compositional baseline featurizers (scale, AAC, ACC, …) + baseline comparison in AAPred #335 (AAC, ACC, and baseline comparison folded into AAPred) — and to avoid reading like df_scales (the letters × scales matrix).
• Fully vectorized backend (byte→scale lookup + one bincount + a single BLAS matmul; no per-sequence Python loop). Correct missing/non-canonical-residue dropping; all-NaN rows for empty spans with a verbose warning. return_df=True for a labeled frame.
• list_parts=None → whole jmd_n + tmd + jmd_c span.

Tests + api-meta green (195); example notebook re-executed with outputs.

🤖 Generated with Claude Code

[breimanntools]

Why do we need this? This is alreday done in CPP.run(split_n_max=1). However, this is indeed a simplificaiton of the process without filtering redudant features. So, it is worht integrateing. Okay from my side

[breimanntools]

This is more a baseline prediction method, or? Should we simply make a baselin predicto and integrate this there? Then we can also add AAC, DPC, CKSAAP and other approaches to get simple baseline predictors? Perhaps in AAPred we have a baseline method. This can load the repreaentations from iFeature +

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.97%. Comparing base (1a152de) to head (a008c3e).
⚠️ Report is 30 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #319      +/-   ##
==========================================
+ Coverage   94.93%   94.97%   +0.03%     
==========================================
  Files         185      187       +2     
  Lines       17883    17994     +111     
  Branches     3038     3051      +13     
==========================================
+ Hits        16978    17089     +111     
+ Misses        598      597       -1     
- Partials      307      308       +1     

Files with missing lines	Coverage Δ
...ature_engineering/_backend/cpp/sequence_feature.py	100.00% <100.00%> (ø)
...aanalysis/feature_engineering/_sequence_feature.py	98.32% <100.00%> (+0.05%)	⬆️

... and 38 files with indirect coverage changes

Components	Coverage Δ
cpp_core	94.95% <ø> (ø)

🚀 New features to boost your workflow:

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