docs(snapshots): note the execution gate covers the single-process run

Clarify that EXECUTION_GATE coordinates threads within one `cargo test`
process (the Linux/macOS jobs, where the parallel-PTY signal flakiness
lives). The Windows job runs under `cargo nextest`, which isolates each
trial in its own process, so the gate is a harmless no-op there and
isolation comes from nextest's process-per-test model instead.

Author: fengmk2

crates/vite_cli_snapshots/tests/cli_snapshots/main.rs

@@ -931,6 +931,14 @@ fn run_case(
 /// else runs while they do. This replaces the old blanket `--test-threads=1`
 /// on Linux: only the few signal-sensitive cases pay for serialization, while
 /// the rest parallelize as they already do on macOS and Windows.
+///
+/// This coordinates threads within a single `cargo test` process, which is the
+/// Linux and macOS snapshot jobs and the only place the parallel-PTY
+/// signal-routing flakiness occurs. The Windows job runs the suite under
+/// `cargo nextest`, which executes each trial in its own process; there the
+/// gate is a no-op, but isolation is stronger for free — a signal-sensitive
+/// case already has its own process, PTY, and process group, which is exactly
+/// what this gate reconstructs for the shared-process case.
 static EXECUTION_GATE: std::sync::RwLock<()> = std::sync::RwLock::new(());
 
 /// Held for a case's whole run: either a shared read lease (parallel) or the