ci: add GitHub Actions workflow (fmt + clippy + test + doc)

[maxine-at-forecast]

Summary

ferrolearn had no PR CI before this — only Dependabot at the org level. This PR adds a strict, parallel-jobs GitHub Actions workflow + the fmt/clippy cleanup needed to land it green from day one.

New workflow (.github/workflows/ci.yml)

Four parallel jobs on every push to main + every PR targeting main:

job	command	gate
fmt	cargo fmt --check	strict rustfmt parity
clippy	cargo clippy --workspace --all-targets -- -D warnings	strict lints
test	cargo test --workspace --release --lib --tests	full suite, release mode (faster for math-heavy paths)
doc	cargo doc --workspace --no-deps --document-private-items	catches syntax-level doc breaks; not strict on warnings yet (see follow-up below)

Architectural choices:

• Independent jobs — fmt / clippy / test / doc all run in parallel so one failure doesn't gate the others; contributors see the full picture in one CI run.
• Concurrency control — cancel-in-progress on the same ref so a fix-up push kills the stale run.
• Permissions hardening — contents: read only. Workflow doesn't push, comment, or deploy.
• Caching — Swatinem/rust-cache@v2 keyed per-job so each artifact pool stays isolated.
• Stable toolchain — no MSRV matrix yet; the workspace declares rust-version = "1.85" but enforcement can come as a separate matrix job if MSRV regressions become a practical concern.

Pre-CI cleanup so it lands green

change	impact
cargo fmt applied across the workspace	66 files touched, ~1500-line diff, pure whitespace. Brings the tree to a rustfmt-clean baseline so future drift gets caught immediately.
8× clippy::needless_range_loop fixed with scoped #[allow]	Math-style matrix index loops (per-class score rows, Laplace smoothing, k-means++ candidate walk) where range-by-index is the natural form. The allow is explicit + per-function so adjacent new code still gets linted. Files: categorical.rs, gmm.rs, lda.rs, gradient_boosting.rs, hist_gradient_boosting.rs, cluster/tests/conformance_sklearn.rs.
1× clippy::approximate_constant (false positive)	3.14 in mlp.rs::test_activate_identity was a placeholder for an Identity-activation passthrough check, not π. Changed to 1.25 with a comment.
1× clippy::unused_imports	Transform not used in ferrolearn-neural/tests/conformance_wave4.rs.
cargo clippy --fix auto-applied	~10 files touched across ferrolearn-preprocess, ferrolearn-tree, ferrolearn-linear, ferrolearn-numerical, ferrolearn-model-sel — clippy::redundant_closure, clippy::derefed_type_is_same_as_origin, clippy::manual_range_patterns, clippy::length_comparison_to_one.

Verification

cargo fmt --check                                                    # clean
cargo clippy --workspace --all-targets -- -D warnings                # clean
cargo test --workspace --release --no-run                            # builds
cargo doc --workspace --no-deps --document-private-items             # builds (1 warning)

Known follow-ups (out of scope)

1. Doc-link cleanup: ~10 broken intra-doc links in ferrolearn-io + ferrolearn-bayes (FerroError::Variant paths that don't resolve). Tighten the doc job to RUSTDOCFLAGS=-D warnings once those land.
2. MSRV enforcement: add a cargo +1.85 check matrix job if MSRV regressions become a practical concern.
3. Coverage / further hardening: deferred.

Diff shape

.github/workflows/ci.yml         | new file (~110 lines)
+
~66 files                        | fmt-only churn (whitespace)
+
~16 files                        | targeted clippy fixes (#[allow] annotations + auto-fixes)

Total: 79 files changed (+1605 / −791), most of it the mechanical fmt churn.

🤖 Generated with Claude Code