[TrapsterDK]

Closes #9588

• elixir_bench (//elixir/rules:defs.bzl): an elixir_test tagged dx-bench and manual. DxBench.run/2 runs Benchee 1.5.1 (mix.lock pinned in elixir/bench/mix.lock, fetched through the existing elixir.mix hub as @dx_benchee) and writes dx-bench.json in ns.
• erlang_bench (//erlang/rules:defs.bzl): a ct_suite tagged dx-bench and manual. dx_bench:run/1,2 runs erlperf 2.3.0 and argparse 2.0.0 (Hex tarballs pinned by digest in erlang/toolchain/pins.bzl) and writes dx-bench.json in ops/s.
• No fuzz macros: no coverage-guided fuzzer exists for the BEAM. Stated in both READMEs.
• Names follow the merged *_bench shape (python_bench, java_bench), not *_benchmark from the issue text. Elixir and Erlang have no <name>_bin because the test rule is the runnable.
• Fixtures: families/beam/{elixir,erlang}/tests/fixtures/bench. Docs: both BEAM READMEs and docs/cli/commands/bench.md.

Tests: ran copies of both fixtures as root packages on Linux with Bazel; both pass and write a valid dx-bench.json. wrapper_contract_tests and the toolchain metadata tests pass. The family module itself does not analyze locally on main (rules_rust_prost is not visible from rules_dx as a non-root module), so the fixtures in families/beam were not run in place. No nested-Bazel or consumer tests were run. docs/support-matrix.md is generated per rule kind and does not list bench macros, so it is unchanged. No CI dispatch: nothing here is platform-specific.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Xut2hwESun1HL2eSVd1d5B