test(metrics): test_gather_metrics depends on another test having run first

[ohdearquant]

What this fixes

ruvector-metrics::tests::test_gather_metrics passes only because another test in the
same process happened to run first.

The metrics are lazy_static, so each one registers with the default Prometheus
registry the first time it is dereferenced. test_gather_metrics dereferences none of
them; it calls gather_metrics() and asserts the output contains "ruvector". Under
cargo test all tests share one process and test_record_search touches two metrics,
so by the time the assertion runs the registry is usually populated. Under a
per-test-process runner the registry is empty and the assertion fails every time.

cargo test        -p ruvector-metrics                                  15 passed
cargo nextest run -p ruvector-metrics                                  14 passed, 1 failed
cargo nextest run -p ruvector-metrics -E 'test(test_gather_metrics)'    0 passed, 1 failed

The workflow runs cargo nextest run, so this is a deterministic failure there, not a
flake.

The change

The test touches a metric itself before gathering, and asserts on that metric's name
instead of the bare "ruvector" substring. It no longer depends on execution order or
on sharing a process with another test.

Verification

• cargo nextest run -p ruvector-metrics: 15 passed.
• cargo nextest run -p ruvector-metrics -E 'test(test_gather_metrics)': 1 passed,
  which is the case that failed before.
• cargo test -p ruvector-metrics: 15 passed.
• cargo clippy -p ruvector-metrics --all-targets -- -D warnings clean; cargo fmt
  applied.

Context

A recent Tests (core-and-rest) run reached its 240-minute cap while still compiling,
so it never got to the test phase and this failure has not been reported there. #784,
#786 and #787 address what keeps that job from finishing; this is one of the results
waiting behind them.