test: add GET response serialization microbench

[houseme]

Summary

• Add an ignored focused microbenchmark for GetObject::serialize_http.
• Split the GET response construction path into empty response, stream body setup, common headers, timestamp formatting, and user metadata cases.
• Keep the benchmark inside ops unit tests so it can exercise the private generated serializer without exposing new public API.

Verification

• cargo fmt --all --check
• git diff --check
• cargo clippy -p s3s --features minio --tests
• cargo test -p s3s --features minio get_object_response_serialization_microbench -- --ignored --nocapture
• cargo test -p s3s --features minio --release get_object_response_serialization_microbench -- --ignored --nocapture

Notes

• This is a diagnostic benchmark only; it does not change production behavior.
• The benchmark reports elapsed ns/op and average response header count. It intentionally avoids adding allocator instrumentation or public benchmarking API.

[houseme]

CI note after the first run:

• rust (stable), rust (1.96.0), coverage, cross-test (macos-latest), cross-test (windows-latest), wasm-test, python, and skip-check passed.
• rust (nightly) failed on existing generated async_trait/#[must_use] Clippy double_must_use diagnostics outside this PR's changed file.
• semver-checks failed on existing public checksum fields in generated DTOs/checksum structs versus the published 0.14.1 baseline; this PR only changes crates/s3s/src/ops/tests.rs.

This PR adds an ignored diagnostic microbenchmark only and does not change public API or production behavior.