perf(helpers): cache or cheapen generate_id per test

[Chemaclass]

Context

bashunit::helper::generate_id (src/helpers.sh:440) fires on every test export and every script load:

• src/runner.sh:29 (per test)
• src/runner.sh:192 (per file load)
• src/runner.sh:317 (per bench file load)

Each call invokes bashunit::helper::normalize_variable_name plus, in parallel mode, bashunit::random_str (src/globals.sh:30) which loops $RANDOM 6 times for chars. Cheap per call, but multiplied across 800+ tests it adds noticeable overhead and one subshell capture per use (\$(bashunit::helper::generate_id ...)).

Proposal

1. Cache normalized script id once per file load — reuse across tests in the same script.
2. For per-test id, append a monotonic counter to the cached script id instead of recomputing:

   : \$((_BASHUNIT_TEST_ID_SEQ += 1))
   export BASHUNIT_CURRENT_TEST_ID=\"\${BASHUNIT_CURRENT_SCRIPT_ID}_\${_BASHUNIT_TEST_ID_SEQ}\"

3. For parallel uniqueness, include \$\$ (already done) — counter + PID guarantees uniqueness without random_str.
4. Eliminate the \$(...) capture at runner.sh:29 by writing directly to the env var.

Acceptance

• generate_id not called per test in hot path (only per script)
• Test ids remain unique within and across parallel workers
• No regression in parallel suite isolation
• All tests green sync + parallel
• Bash 3.0+ compat preserved
• Benchmark before/after on tests/unit/ posted in PR

Risk

• Test IDs used in temp_file/temp_dir naming (src/globals.sh:41+). Verify no consumer expects random suffix.