perf(helpers): inline normalize+random in generate_id; drop grep fork

[Chemaclass]

Summary

• bashunit::helper::normalize_variable_name now tests the first-character identifier rule with a pure-bash case glob instead of piping into grep. Drops one fork per call across all callers (snapshot, test_doubles, generate_id).
• bashunit::helper::generate_id inlines the normalization and random-suffix logic so it no longer subshells into normalize_variable_name or random_str. Two forks saved per call.
• generate_id is invoked once per test and once per file load. On the current 800-test suite that's ~1600 forks eliminated.

Why not a counter-based ID?

• The issue proposed a monotonic counter to replace random_str, but generate_id is invoked via $(...) at all call sites — counter increments inside a subshell don't persist to the parent. Inlining random_str (cheap: 6 $RANDOM indexes, no fork) gets the same end result without touching call sites.

Test plan

• New tests in tests/unit/helpers_test.sh:
  • test_generate_id_uses_pid_suffix_when_not_parallel
  • test_generate_id_appends_random_suffix_when_parallel
  • test_generate_id_sanitizes_basename
• Existing test_normalize_variable_name covers all edge cases (empty string, leading digit, leading underscore, all-special-chars)
• ./bashunit tests/unit/ green (817 passed)
• ./bashunit --parallel tests/ green (1052 passed, 46s)
• make sa clean
• make lint clean
• Bash 3.0+ compat preserved

Closes #663