chore: pin Rust toolchain to 1.94.0 and declare MSRV

[orrinfrazier]

Closes #18

Summary

• Pins the Rust toolchain to 1.94.0 via a new root rust-toolchain.toml and declares an MSRV of 1.94 via [workspace.package].rust-version, inherited by all 30 member crates.
• Updates ci.yml, release.yml, and deny.yml so non-Guix CI and the public release binaries stop floating on whatever stable is current at run time.
• Implements issues/S06-1-msrv-toolchain-pin.md.

Why

Without a pinned toolchain, a new stable release can silently break CI or change release-binary output, reproducible builds can't be guaranteed for the non-Guix matrix, and contributors/distros have no declared minimum supported Rust version.

Changes

• rust-toolchain.toml (new): channel = "1.94.0", components rustfmt + clippy.
• [workspace.package].rust-version = "1.94" + rust-version.workspace = true on every member crate (scripted across 30 manifests). cargo metadata reports rust_version = 1.94 for all 30 members.
• Workflows: every toolchain: stable → toolchain: 1.94.0 (with a sync comment). dtolnay/rust-toolchain does not read rust-toolchain.toml, so the version is named explicitly in each workflow. doc.yml and fuzz.yml remain on nightly intentionally (cargo-doc features / cargo-fuzz).
• MSRV policy: set equal to the pin (latest-stable-only), documented inline — not a fabricated lower bound (no MSRV-verification CI job exists to back one).
• constants/src/lib.rs: a std-only #[cfg(test)] mod toolchain_pin that locks the invariants — globs all .github/workflows/*.yml (asserting no toolchain: stable, pinned version everywhere, nightly workflows allowlisted) and iterates all workspace members (asserting each inherits the MSRV). Globbing/iterating rather than sampling means a future job left on stable or a new crate missing inheritance is caught.

How to test

cargo test -p cuprate-constants toolchain_pin     # 4/4 green
cargo metadata --format-version 1 --no-deps        # every member rust_version = 1.94
cargo fmt -p cuprate-constants --check
cargo clippy -p cuprate-constants --all-targets --all-features -- -D warnings

The drift guard was verified negatively: injecting a workflow on toolchain: stable and removing inheritance from a member each make the corresponding test fail.

Review notes

Reviewed via 3-way cross-family consensus (Claude + Gemini + Codex, opus adjudicator). Consensus strengthened the drift-guard test from a hardcoded 3-file/3-crate sample to globbing all workflows + iterating all members. Rejected with rationale: pinning doc.yml/fuzz.yml to 1.94.0 (nightly is required there), removing/faking a lower MSRV, and adding the wasm32 target to rust-toolchain.toml (CI installs it per-job; global cost not justified).