Document the around_test instrumentation hook

[KyleOps]

Companion documentation for inferno-core#797, which adds TestRunner#around_test. Raised at @karlnaden's request on that PR, which suggested either the Debugging page or a new page under Advanced Features. I went with a new page, and linked to it from Debugging, because the two cover different concerns: Debugging is about investigating a single test interactively while writing it, this is about observing runs in a deployed instance.

What the page covers

• Why instrumentation applied around the runner is not useful for a long run: Inferno executes a whole run in one call stack, so a tracing library that instruments the background worker sees one long-lived operation rather than one per test.
• How to override the hook, with a small logging/timing example.
• The contract an override has to honour: call super exactly once, return that value unchanged, let exceptions propagate. Each with the consequence of getting it wrong, since all three fail silently.
• A worked OpenTelemetry example producing one bounded trace per test.
• Where to put the override so it loads in every process that runs tests.

Two things worth flagging for review

The span-naming section is the part I would most like checked. The example deliberately gives the span a constant name and puts identity in attributes. This is not a style preference: tracing backends treat the span name as a low-cardinality dimension and generate metrics keyed on it, so naming a span after the test produces one metric series per test, per suite, per suite version. On our deployment of a large test kit that was over 1,400 distinct span names, and every one of those series is useless by construction, because a test emits a single span per run and rate() or increase() over one observation returns zero. An earlier draft of the inferno-core PR had the interpolated form, and it seemed worth documenting the trap rather than shipping an example that leads people into it.

The page notes that the hook wraps all of #run_test, which includes persisting the result along with its messages and requests. A duration measured with the hook is the wall time a user waited, which is usually what you want, but it is not purely test execution. Flagged because it surprised us: for request-heavy tests that persistence is a meaningful share of the elapsed time.

Also

docs/advanced-test-features/index.md gets an entry for the new page, and docs/getting-started/debugging.md gets a short closing section pointing at it.

Happy to fold this into the Debugging page instead, or to cut the span-naming section back, if you would rather the page stayed shorter.

[KyleOps]

"Keep the test out of the span name" section might be too opinionated but it is a bit of a gotcha so I think it is worth it.