fix(bench): fail the run when a benchmark file fails to load

[Chemaclass]

🤔 Background

Related #1303

bashunit bench printed a file-level failure and then exited 0. Same scenario, both subcommands — one good file plus one whose set_up_before_script fails:

$ bashunit bench benchmarks/
✗ Error: Set up before script
bench_works    1    1    6
EXIT=0            <- bench

$ bashunit tests/
EXIT=1            <- test

A syntax error was worse: every bench_ function after it vanished from the table and nothing was reported as failed at all.

💡 Changes

• load_bench_files checks the sourcing status and matches the captured stderr for syntax error / unexpected EOF, as the test loop does — bash reports a syntax error and carries on with a zero status, so the status alone does not catch it. case, not grep, to stay fork-free
• cmd_bench consults the failure counter. It exited non-zero only for "No benchmarks found" and the baseline gate, so the existing set_up_before_script failure path could not affect the verdict — it only looked correct when the failure left nothing to run and "No benchmarks found" fired instead
• A benchmark that merely returns non-zero still exits 0: a benchmark measures time and has no assertion concept

🔍 How it was found

Applying the pathological-fixture sweep from #1295/#1297/#1299/#1301 to bench, which has its own runner and reports and is far less exercised than test.