perf(state): avoid base64 forks for empty title and hook message in export_subshell_context #762

Part of #761.

Problem

Every test run encodes its subshell context once per test in bashunit::state::export_subshell_context (src/state.sh:231-259). It always forks three echo -n ... | base64 pipelines (src/state.sh:239-246) for _BASHUNIT_TEST_OUTPUT, _BASHUNIT_TEST_TITLE, and _BASHUNIT_TEST_HOOK_MESSAGE, even though title and hook message are empty for almost every test. On the decode side, bashunit::helper::decode_base64 (src/helpers.sh:149) already has a fast path for the _BASHUNIT_EMPTY_ marker, but a plain empty string still forks printf | base64 -d, and bashunit::runner::decode_subshell_output (src/runner.sh:1110) runs once per test.

Measured cost: each pipeline is ~1.6ms (2 forks) on macOS bash 3.2. That is ~4-5ms per test, ~5s across the ~1160-test suite, and it compounds inside the 258 nested ./bashunit runs in the acceptance suite.

Task

1. In export_subshell_context, only fork base64 when the source variable is non-empty; emit an empty string (or the existing _BASHUNIT_EMPTY_ marker, whichever keeps the wire format consistent — check how ##TEST_TITLE=/##TEST_HOOK_MESSAGE= are parsed downstream in src/runner.sh and src/parallel.sh) for empty values. Note echo -n "" | base64 already produces an empty string, so guarding with [ -n "$var" ] should be wire-compatible; verify against the parallel .result file path too (bashunit::runner::parse_result_parallel, src/runner.sh:1441).
2. In decode_base64, add an early return for empty input so "" never forks.
3. Audit callers of decode_base64 for other per-test empty decodes.

Tests first (TDD)

• Unit tests for encode/decode round-trip: empty string, plain string, multiline string, string with ANSI codes, UTF-8. Follow patterns in tests/unit/helpers_test.sh / existing state tests.
• Regression: a test with a # @test_title annotation and a hook message still renders correctly (sequential AND --parallel), since those exercise the non-empty paths.

Constraints

• Bash 3.0+ only: no declare -A, no [[ ]], no ${var,,}, no negative array indexing, no &>>.
• Wire format of ##ASSERTIONS_*##TEST_* context string must remain parseable by the existing extractors; do not change field order.
• Keep the Alpine base64 -w 0 branch behavior (src/state.sh:237-241).

Acceptance criteria

• ./bashunit tests/ and ./bashunit --parallel tests/ pass.
• make sa, make lint pass; shfmt -w . produces no diff.
• Fork census (see tracking issue) shows base64 count drops from ~4/test to ~2/test for tests without title/hook message.
• Before/after timing of the 100-noop benchmark reported in the PR.

Line references valid at commit 4d80e7c.