refactor: dissolve benches/examples crates into host crates

[mfw78]

Closes #618

Summary

Dissolved crates/benches and crates/examples into their host crates and swapped the codspeed shim for plain criterion.

Bench homing: address, bmt_bench, encryption_bench, primitives, proofs to nectar-primitives; latency_bench to nectar-file; mantaray_bench to nectar-mantaray; verify to nectar-postage; sign and upload_pipeline to nectar-postage-issuer. Example homing: basic_usage and builder_patterns to nectar-primitives; dump_deployments to nectar-contracts; stamp_pipeline to nectar-postage-issuer; roam_between_machines to nectar-postage-usage. All fifteen files moved with git mv so history follows.

Feature gating uses required-features, not a self dev-dependency. A self dev-dependency (the tokio pattern) would have unified the extra feature into every build of the host package that activates dev-dependencies, silently turning the default-feature CI matrix rows into feature-on rows.

crates/benches and crates/examples are removed as workspace members; their Cargo.toml and src/lib.rs are deleted.

Testing

All commands run inside nix develop --command at 2c18d09c (branch refactor/618-dissolve-bench-crates, based on origin/main @ 1ea486b3). Toolchain: rustc 1.94.0, which equals the declared MSRV (.github/scripts/declared-msrv.sh -> 1.94), so the build below doubles as the MSRV lane's cargo check --workspace --all-targets --locked.

Workspace member count (cargo metadata --no-deps): 23 before -> 21 after. nectar-benches and nectar-examples are the two removed members.

Build and target relocation

• cargo build --workspace --all-targets --locked: pass. Every moved bench and example builds in its new host crate.
• cargo bench --workspace --locked --no-run: pass (default-feature benches).
• cargo bench --locked -p nectar-postage --features parallel --no-run: pass (verify).
• cargo bench --locked -p nectar-mantaray --features hazmat: pass (mantaray_bench).
• All ten benches and five examples compile in their new hosts.
• All 14 clippy rows in lint.yml pass with -D warnings.
• cargo fmt, cargo deny check (advisories/bans/sources/licenses), and cargo-machete are clean.
• nextest is 573/573 on the seven touched crates.
• cargo run -p nectar-contracts --example dump_deployments emits the JSON upstream-addresses.yml consumes.
• nostd.yml uses plain cargo check; no bench target can leak onto the bare-metal lane.

AI Assistance

Implemented by claude-opus-5. Red-teamed by claude-opus-5. PR description by claude-sonnet-5.

[mfw78]

Correction to the member count, for the record, since stage 2's gate is that the final count is written down and correct.

The brief I gave said 22 members before and 20 after. That was wrong in both directions, and the red-team caught it. The workspace declares members = ["crates/*", "crates/primitives/examples/wasm-demo"], so the member count is the number of directories under crates/ plus one: crates/primitives/examples/wasm-demo is an explicitly listed member that a directory count misses.

The true figures are 23 members before this pull request and 21 after. The directory count under crates/ did go 22 to 20, which is where my number came from, but that is not the member count.

This matters beyond bookkeeping: stage 2's exit gate in docs/PRODUCTION-PLAN.md is that cargo metadata reports the final member count and it is written down, and #713 and #716 both enumerate the tree for the release. Anything quoting 20 should quote 21.