perf: isolate benchmark and e2e dependencies

[NickNYU]

Summary

• move the four Criterion benchmarks into a dedicated agentenv-benchmarks workspace package
• move the Docker/MinIO OSS snapshot test into agentenv-e2e-tests
• keep the existing Makefile entry points while removing Criterion and test-support from the root package dev-dependencies
• update internal documentation for the relocated OSS test

Part of #8 (P1).

Why

Cargo activates a package's dev-dependencies for library test builds even when the test target does not use them. Keeping Criterion and agentenv-test-support on the root package caused cargo test -p agentenv --lib --no-run to compile benchmark tooling, AWS SDK crates, and testcontainers.

Dedicated workspace packages preserve the benchmark/e2e targets while moving those costs behind explicit package selection.

Measured result

Same isolated VM for both runs: Ubuntu 22.04 ARM64, 8 vCPU, 16 GiB RAM, rustc 1.97.1. Commands used a fresh target directory and a warm Cargo source cache.

cargo test -p agentenv --lib --no-run --locked	Before	After	Change
wall time	243.41s	221.34s	-22.07s (-9.1%)
target size	7.4 GiB	6.6 GiB	about -0.8 GiB
compiler artifacts	749	649	-100 (-13.4%)

The compiler-artifact gate found all six P1 dependencies before the change and none afterward:

• criterion
• agentenv-test-support
• aws-sdk-s3
• aws-config
• testcontainers
• testcontainers-modules

The ARM64 timings are screening data rather than a replacement for the issue's x86_64 baseline; the dependency isolation result itself comes from Cargo's compiled artifact stream and dependency tree.

Verification

• cargo fmt --all -- --check
• cargo test -p agentenv --lib --locked — 633 passed, 0 failed, 2 ignored
• cargo test -p agentenv-e2e-tests --test snapshot_oss_e2e_test --no-run --locked
• cargo test -p agentenv-benchmarks --benches --no-run --locked — 4 executables
• cargo clippy -p agentenv -p agentenv-benchmarks -p agentenv-e2e-tests --all-targets --locked -- -D warnings
• root dependency-tree gate — 0 forbidden dependencies
• Makefile dry-run for benchmark and OSS e2e entry points

Not run

• Docker/MinIO-backed ignored e2e tests at runtime
• KVM/capability-dependent benchmark execution

[github-actions]

🔍 OpenCodeReview found 1 issue(s) in this PR.

• ✅ Successfully posted inline: 1 comment(s)

[github-actions]

This package defines only benchmark targets, so these crates are benchmark-only dependencies. Keeping them under [dependencies] exposes them as normal package dependencies and broadens normal dependency/feature resolution unnecessarily. Move the benchmark dependency list to [dev-dependencies], which Cargo makes available to bench targets.

Suggestion:

Suggested change

	[dependencies]
	agentenv = { path = "../.." }
	anyhow = "1.0.100"
	criterion = { version = "0.8", features = ["async_tokio"] }
	[dev-dependencies]
	agentenv = { path = "../.." }
	anyhow = "1.0.100"
	criterion = { version = "0.8", features = ["async_tokio"] }

[LSX-s-Software]

Thanks for contributing. Could you please fix the comment posted by the AI reviewer?