docs(book): walkthroughs for 30 aprender-core public modules (BOOK-CLOSEOUT-001 follow-up)

[noahgift]

Summary

Replace 20-line lib-module stubs with full walkthroughs for 30
high-value aprender-core public modules. Each chapter now has a
module summary, key types table, two runnable usage patterns, and
see-also cross-links to related modules.

Modules covered (30):
linear_model, classification, cluster, ensemble, traits,
primitives, format, nn, optim, loss, metrics, models, autograd,
decomposition, time_series, text, audio, graph, gnn, bayesian,
tree, glm, regularization, model_selection, data, calibration,
preprocessing, serialization, interpret, online.

Where the user task referenced names that aren't top-level
modules in the actual crate (svm, naive_bayes,
neighbors, quantize, linear_regression), the
walkthrough was authored on the closest real module —
svm / naive_bayes / neighbors are documented inside
classification.md, linear_regression inside
linear_model.md, and quantize content folded into
preprocessing.md (with format-level quantization referenced
in format.md). Substitute additional modules
(preprocessing, serialization, interpret, online)
were authored so the batch lands a full 30 walkthroughs.

LOC added: ~2,068 across 30 files (averaging ~70 net new
lines per chapter, from ~20 to ~85).

Fixed default stub examples that referenced non-existent
symbols:

• aprender::traits::{Estimator, Predictor} → Estimator, Transformer, UnsupervisedEstimator
• aprender::ensemble::RandomForest → MixtureOfExperts, SoftmaxGating, MoeConfig
• aprender::loss::{MeanSquaredError, CrossEntropy} → MSELoss, MAELoss, HuberLoss, Loss
• aprender::metrics::{accuracy, f1_score} → split between top-level r_squared/mse/mae/rmse and metrics::classification::{accuracy, f1_score, Average}
• aprender::format::{Reader, Writer} → ModelCard, AprConverter, ConvertOptions
• aprender::autograd::{Variable, backward} → Tensor, no_grad
• aprender::decomposition::PCA → ICA (PCA actually lives in preprocessing)
• aprender::text::ChatTemplate → ChatTemplateEngine, ChatMessage
• aprender::audio::{load_wav, MelSpectrogram} → MelConfig, MelFilterbank, validate_audio
• aprender::graph::{dijkstra, pagerank} → Graph, GraphCentrality (these are trait methods, not free functions)
• aprender::gnn::GraphConvNetwork → GCNConv, GATConv, GINConv
• aprender::glm::{PoissonGLM, GammaGLM} → GLM, Family, Link
• aprender::regularization::{L1, L2, Dropout} → Mixup, LabelSmoothing, CutMix, StochasticDepth
• aprender::data::DataLoader → DataFrame, ColumnStats

Test plan

• bash scripts/check_book_lib_example_block.sh — PASS (all 69 lib chapters have a rust fenced block)
• bash scripts/check_book_lib_parity.sh — PASS (69/69 modules have chapters)
• bash scripts/check_book_linkcheck.sh — PASS (0 broken file links)
• mdbook build — clean (only pre-existing search-index-size warning)
• No placeholder strings (TODO / TBD / WIP / coming soon / under construction) in any of the 30 authored files
• Every rust fenced block in the 30 authored files uses real public symbols verified against crates/aprender-core/src/<mod>/

🤖 Generated with Claude Code