Run teardown / a capture hook even when a test times out

## Problem

When a `timeout` fires, the test subprocess is hard-killed and per-test **teardown does not run** (as documented). For a suite driving a shared external backend, a timed-out test can leave that backend in a bad state, cascading failures into every subsequent test. There's also no hook to capture context (logs/state) at the moment of failure or timeout.

## Proposal

1. On timeout, send a cancellation/grace signal and run the registered `ctx.teardown(...)` within a **bounded grace window** before the hard-kill.
2. An optional per-test **on-failure / on-timeout hook** (a closure) that runs to capture diagnostics and attach them to that case's report.

```rust
#[testcase(timeout = Duration::from_secs(30), teardown_grace = Duration::from_secs(5))]
async fn t(ctx: Arc) -> Result<(), rigtest::Error> { /* ... */ Ok(()) }
```

## Why

Guarantees cleanup so one slow/wedged test can't poison siblings that share a backend, and gives a capture point that actually runs when a test dies. Complements #42 (failure capsule) — that packages artifacts; this ensures teardown + a capture point run on timeout. Bounded grace so a wedged teardown can't itself hang.

The issue names the test subprocess timeout path, the registered ctx.teardown hook, and the proposed per-test capture hook. Start by tracing where timeout handling hard-kills the subprocess and how teardown is registered; done means bounded cleanup and diagnostic capture are specified and verified without allowing a wedged teardown to hang.

由索引模型根據 Issue 內容生成。