docs: bring stale claims in line with what shipped

[Burton-David]

Documentation hadn't caught up with Phases 2-5. Pulled out of the audit as the highest-value cluster — seven concrete corrections, no code change.

File	Problem	Fix
serving/README.md	Opens with "the answer is: not really" — directly contradicts the Phase 5 ADR (same file, lower down) showing Go won throughput/latency/memory	Rewrite intro to match measurement
README.md:67	Claims BPR is "~30×" faster	Bump to "~51×" to match the ADR + benchmarks/profile.md
README.md:122	Algorithm table called BPR "numpy SGD"	"Rust+PyO3 kernel; Python fallback"
docs/index.md	"Five algorithms out of the box" + pip install recommender-systems	List all ten algorithms; source-install recipe with the Rust toolchain prereq
docs/choosing_an_algorithm.md	BPR "when it isn't" still mentioned the numpy SGD bottleneck (gone since Phase 2); k-NN/SVD rows said "dense"	Reframed BPR to ALS-as-alternative; updated data-source columns to reflect Phase 3's sparse rewrite
ROADMAP.md	Pre-Phase-2: book recommender listed as in-progress (shipped); #77 sparse work listed as next (shipped); Rust kernel and serving experiment entirely absent	Rewrote Shipped to reflect reality; Next is release + ContentBased-on-sparse
src/recommender_systems/books.py	Module docstring used "showcase" — the one remaining instance, surfaced into api.md via mkdocstrings	"recommender"
docs/evolution/02-rust-kernel-bpr.md:73	Pointed at tests/test_fit_speed.py	tests/benchmarks/test_fit_speed.py

Verification

• ruff check / ruff format --check / mypy clean
• pytest — 137 passed, 1 skipped (kernel without build), 7 deselected
• mkdocs build --strict — 0 warnings

[JohnJacob-coder]

Checked each correction against main. The ~51x and the Rust-kernel BPR row match the Phase 2 ADR and benchmarks/profile.md; the k-NN/SVD doc rows match the Phase 3 sparse rewrite; the serving README intro no longer contradicts its own ADR; and the books.py docstring was the last 'showcase' anywhere (grep across README/docs/src/serving is clean now). ruff/mypy/pytest green, mkdocs --strict builds with no warnings. Good catch on the doc drift.