chore(toolchain): bump pin from 1.94.0 to 1.94.1 (matches sibling repos)

[AdaWorldAPI]

Summary

Bumps rust-toolchain.toml from 1.94.0 to 1.94.1 (latest 1.94 patch, 2026-03-25). Matches the same pin landing on AdaWorldAPI/ndarray, AdaWorldAPI/smb-office-rs, and AdaWorldAPI/MedCare-rs so all sibling repos work against the same compiler.

Why

Clippy 1.95 turned several previously-safe patterns into denied lints (e.g. clippy::unnecessary_sort_by) without value sufficient to justify the churn. Pinning to 1.94.1 keeps lance-graph on the last-good 1.94 patch.

The existing pin was 1.94.0; bumping to 1.94.1 is a 1-line change. Comment block updated to record the bump policy ("never auto-track stable; review lint changes when bumping").

Note on CI workflows

Some workflows (rust-test.yml, style.yml, rust-publish.yml, release.yml) use actions-rust-lang/setup-rust-toolchain@v1 with explicit toolchain: stable inputs that OVERRIDE the toml pin. This PR does not touch those — the toml bump helps local dev environments track the same version, but CI workflow alignment is a follow-up. Open question for that follow-up: do the matrix-style jobs in rust-test.yml (toolchain: [stable]) intend cross-toolchain testing, or are they accidental override?

Cross-ref

• MedCare-rs PR feat: Qwen 3.5 9B Opus v2 + Qwen 2.5 3B codebooks (7 models total) Qwen3.5-9B (Jackrong/Opus-Reasoning-Distilled-v2, Q8_0, 9.53 GB): 64×1024 centroids + 64² table + 4096 assignments Assignment skew: min=25, max=195 (Opus distillation effect) Distance table min=75 (wider range than other models) Qwen2.5-3B (bartowski Q8_0, 3.29 GB): 64×256 centroids + 64² table + 2048 assignments 7 models total: Jina, BGE-M3, GPT-2, MiniLM, reader-LM, Qwen2.5-3B, Qwen3.5-9B Total codebook storage: ~880 KB for ALL models. https://claude.ai/code/session_01ChLvBfpJS8dQhHxRD4pYNp #89 — same pin pattern in the consumer with the full failure-mode story.

https://claude.ai/code/session_01SbYsmmbPf9YQuYbHZN52Zh

---

Generated by Claude Code