test(spurctld): replace metrics_server unit tests with meaningful handler tests

[shiv-tyagi]

Ref: #217

The current tests in crates/spurctld/src/metrics_server.rs (leader_returns_slurm_content_type, leader_returns_openmetrics_content_type, stub_nodes_endpoint_returns_200_on_leader, follower_returns_503) test a #[cfg(test)]-only helper function leader_metrics_response that doesn't exist in production code. They verify that axum's tuple-to-response conversion returns the status code and content-type it was given, e.g.:

let response = leader_metrics_response(true, format, body);
assert_eq!(response.status(), StatusCode::OK);

This is testing axum, not Spur. The leader_metrics_response helper is just:

fn leader_metrics_response(is_leader: bool, format: MetricsExpositionFormat, body: String) -> Response {
    if !is_leader { return not_leader_response(); }
    metrics_exposition_response(format, body)
}

Both not_leader_response and metrics_exposition_response are trivial one-liner tuple constructions. There's no Spur logic under test.

What to do

• Remove leader_metrics_response and the four tests that use it.
• Replace with tests that exercise the actual axum handlers (metrics_jobs, metrics_nodes, etc.) using axum::test::TestClient or by constructing a Router with test state. These would cover the real flow: Raft leader check, snapshot collection, encoding, and correct response format.
• If full handler tests are too heavy for this PR, even just removing the current tests and the dead helper is a net improvement since they add maintenance cost without catching real bugs.