CRITICAL: Test name extraction via CLI args is unreliable

[siphonite]

Problem

The orchestrator attempts to infer the test name by parsing std::env::args(). This logic is fragile and prone to failure.

Code reference:
https://github.com/siphonite/first/blob/main/src/orchestrator.rs#L328-L348

fn extract_test_name() -> Option<String> {
    // ...
    // Skip the executable path, look for something that looks like a test name
    for arg in args.iter().skip(1) {
        // ... (skipping flags)
        return Some(arg.clone());
    }
    None
}

This fails when:

• No explicit test filter is provided.
• Flags are reordered or complex.
• Cargo behavior changes (e.g., passing extra arguments).

If extraction fails, child processes may run all tests (default behavior when no test name is given), causing recursive spawning or cross-test interference.

Impact

• Exponential test spawning: If a child process meant to run a specific test instead runs all tests, it will spawn its own children, leading to a fork bomb.
• Cross-test contamination: Tests interfering with each other's state.
• Potential infinite recursion: The orchestrator spawning itself endlessly.

Acceptance Criteria

• Test identity must be unambiguous and explicit.
• Preferred: mandatory FIRST_TEST_NAME environment variable set by the orchestrator when spawning children.
• Or: robust mechanism (not CLI parsing) to identify the current test.
• Orchestrator must refuse to run if test identity is unknown.