docs: benchmark example invocations in benchmarks.md omit required cargo features

[ohdearquant]

Problem

The "End-to-end example benchmarks" section of docs/benchmarks.md gives copy-pasteable cargo run commands, but the feature flags do not match what crates/inference/Cargo.toml requires for these examples, so following the doc as written fails to run:

cargo run -p lattice-inference --example bench_embedding --release
cargo run -p lattice-inference --example bench_metal --release --features metal-gpu
cargo run -p lattice-inference --example bench_concurrent --release

• bench_embedding requires the f16 feature (the doc passes none)
• bench_metal requires f16 and metal-gpu (the doc passes only metal-gpu)
• bench_concurrent requires f16 and metal-gpu (the doc passes none)

Evidence

Reported from a user walkthrough at https://raw.githubusercontent.com/mondweep/vibe-cast/lattice-exploration/learnings.md (the bench_metal command was missing f16).

• docs/benchmarks.md:60,63,66 hold the three commands above.
• crates/inference/Cargo.toml sets required-features = ["f16"] for the bench_embedding example and required-features = ["f16", "metal-gpu"] for the bench_metal and bench_concurrent examples.

Suggested fix

Update the three commands in docs/benchmarks.md to pass the full required feature set (--features f16 and --features f16,metal-gpu). Docs-only change.

[ohdearquant]

Comment authored by an AI agent on behalf of @ohdearquant.

Resolved on main: docs/benchmarks.md:60-66 lists the required --features flags for all three benchmark examples, matching Cargo.toml required-features. Verified at source.