fix(pipeline): run compound command stages concurrently

[hartsock]

Found during practical testing of compound commands.

A compound command used as a non-final pipeline stage runs inline to completion instead of being spawned, so the stage that would drain its output is never started. Once it writes more than one pipe buffer (64 KiB on Linux/macOS) it blocks forever.

# completes (62 KiB)
i=0; while [ $i -lt 2000 ]; do printf '%s\n' 0123456789012345678901234567890; i=$((i+1)); done | cat
# hangs (65 KiB)
i=0; while [ $i -lt 2100 ]; do printf '%s\n' 0123456789012345678901234567890; i=$((i+1)); done | cat

bash completes both. The common shape is ... | while read x; do ...; done | sort | head.

Spawn the command onto a blocking task when it owns its shell — the same shape execute_via_builtin_in_owned_shell already uses for builtin stages. The parent-shell case (single-command pipeline, or the last stage under lastpipe) still runs inline, preserving its side effects.

The owned-shell path already executed against Box::new(shell.clone()), so stage side effects were discarded before this change as well; only when the stage runs changes, not which shell it mutates.

Function and ExtendedTest share the inline shape but write nothing to stdout, so they cannot fill a pipe — left unchanged.

Adds one compat case (128 KiB through a while stage). Compat suite: 1773 passed / 0 failed, against 1772 / 0 on main — exactly the added case, known-failures unchanged at 380.

Tested in our downstream fork.

[hartsock]

Found alongside #1184 and #1244 during practical testing of brush embedded as a shell library.

[github-actions]

Performance Benchmark Report

Benchmark name	Baseline (μs)	Test/PR (μs)	Delta (μs)	Delta %
clone_shell_object	17.66 μs	17.48 μs	-0.18 μs	🟢 -1.03%
eval_arithmetic	0.15 μs	0.15 μs	-0.00 μs	⚪ Unchanged
expand_one_string	1.69 μs	1.62 μs	-0.07 μs	⚪ Unchanged
for_loop	32.18 μs	31.68 μs	-0.50 μs	⚪ Unchanged
full_peg_complex	57.08 μs	56.80 μs	-0.28 μs	⚪ Unchanged
full_peg_for_loop	6.17 μs	6.27 μs	0.09 μs	🟠 +1.52%
full_peg_nested_expansions	16.11 μs	16.17 μs	0.07 μs	⚪ Unchanged
full_peg_pipeline	4.24 μs	4.26 μs	0.02 μs	⚪ Unchanged
full_peg_simple	1.78 μs	1.80 μs	0.02 μs	⚪ Unchanged
function_call	3.46 μs	3.42 μs	-0.03 μs	⚪ Unchanged
instantiate_shell	54.74 μs	55.07 μs	0.34 μs	⚪ Unchanged
instantiate_shell_with_init_scripts	27053.55 μs	26861.39 μs	-192.16 μs	⚪ Unchanged
parse_peg_bash_completion	2111.45 μs	2108.34 μs	-3.12 μs	⚪ Unchanged
parse_peg_complex	20.71 μs	20.82 μs	0.11 μs	⚪ Unchanged
parse_peg_for_loop	2.01 μs	2.04 μs	0.04 μs	🟠 +1.84%
parse_peg_pipeline	2.05 μs	2.11 μs	0.06 μs	🟠 +2.83%
parse_peg_simple	1.07 μs	1.08 μs	0.01 μs	⚪ Unchanged
run_echo_builtin_command	16.73 μs	16.64 μs	-0.10 μs	⚪ Unchanged
tokenize_sample_script	3.42 μs	3.35 μs	-0.06 μs	🟢 -1.79%

Code Coverage Report: Only Changed Files listed

Package	Base Coverage	New Coverage	Difference
Overall Coverage	🟢 27.23%	🟢 27.23%	⚪ 0%

Minimum allowed coverage is 20%, this run produced 27.23%
Maximum allowed coverage difference is -5%, this run produced 0%
3.2% |
| brush-test-harness/src/execution.rs | 🟢 80.2% | 🟢 83.17% | 🟢 2.97% |
| brush-test-harness/src/reporting.rs | 🔴 29.93% | 🔴 46.48% | 🟢 16.55% |
| Overall Coverage | 🟢 75.49% | 🟢 75.76% | 🟢 0.27% |

Minimum allowed coverage is 70%, this run produced 75.76%
Maximum allowed coverage difference is -5%, this run produced 0.27%

Test Summary: bash-completion test suite

Outcome	Count	Percentage
✅ Pass	1582	75.01
❗️ Error	17	0.81
❌ Fail	156	7.40
⏩ Skip	339	16.07
❎ Expected Fail	13	0.62
✔️ Unexpected Pass	2	0.09
📊 Total	2109	100.00