[Bug]: Test Infrastructure Fragility: data-driven AprilTag tests have silent failures and panic-on-error paths

🐛 Describe the bug

The data-driven test infrastructure in crates/kornia-apriltag/tests/data_driven.rs has multiple fragility points that can cause silent failures or misleading test results:

1. partial_cmp().unwrap() panics on NaN (line 170) — If the detector produces NaN coordinates, f32::partial_cmp returns None and .unwrap() crashes with an unhelpful "called unwrap() on a None value" instead of reporting the actual geometric issue.

2. expect() panics instead of returning Err (line 225) — The run_data_driven_tests function returns Result<()>, but the Edge test arm uses .expect("Edge test must have min_detections") which panics instead of propagating an error through ?. A manifest authoring mistake produces a panic, not a diagnostic.

3. Golden tests silently pass with zero validation — If a Golden test case has neither expect_ids nor expect_detections, both if let Some(...) branches are skipped. The detector runs, but nothing is asserted. The test reports success while validating nothing.

4. No finite-value validation on geometric coordinates — NaN or Infinity in manifest coordinates or detector output makes assert!((a - b).abs() <= tol) comparisons meaningless (NaN comparisons always return false, but the assertion message is misleading).

🔄 Steps to Reproduce

1. Clone the repo and check out main

2. Add a Golden test case to crates/kornia-apriltag/tests/data/manifest.json with no expect_ids and no expect_detections — it silently passes

3. Add an Edge test case without min_detections — it panics instead of returning an error

4. Inspect line 170: da.partial_cmp(&db).unwrap() — any NaN from the detector causes an opaque panic

💻 Minimal Code Example

// Fragility 1: Panic on NaN (line 170)
let da: f32 = f32::NAN;
let db: f32 = 1.0;
da.partial_cmp(&db).unwrap(); // panics: "called `Option::unwrap()` on a `None` value"
// Fix: da.total_cmp(&db)  — total ordering, no panic, already used in quad.rs

// Fragility 2: Panic instead of Err (line 225)
let min_detections: Option<usize> = None;
min_detections.expect("Edge test must have min_detections"); // panics
// Fix: min_detections.ok_or_else(|| format!(...))?

// Fragility 3: Silent pass — this Golden arm does nothing
let expect_ids: Option<Vec<usize>> = None;
let expect_detections: Option<Vec<ExpectedDetection>> = None;
if let Some(ref ids) = expect_ids { /* skipped */ }
if let Some(ref dets) = expect_detections { /* skipped */ }
// Test passes having validated zero assertions

✅ Expected behavior

~ All test code paths either validate something meaningful or return a clear error ~ Manifest authoring mistakes are caught with descriptive error messages, not panics ~ NaN/Infinity in coordinates is detected and reported, not silently swallowed ~ A Golden test that validates nothing is treated as a configuration error

❌ Actual behavior

~ .unwrap() on partial_cmp panics with an opaque message when NaN is encountered ~ .expect() on min_detections panics instead of returning Err through the Result return type ~ Golden tests with no expectations silently pass — the detector runs but nothing is asserted ~ Non-finite coordinates produce meaningless comparison results with misleading assertion messages

🔧 Environment

- kornia-rs version: 0.1.11-rc.1
- Rust version (`rustc -V`): 1.92.0
- Cargo version (`cargo -V`): 1.92.0
- OS (e.g., Linux, macOS, Windows): macOS
- Target architecture (if cross-compiling): aarch64
- Python version (if using Python bindings):  N/A

📝 Additional context

~ The crate already uses f32::total_cmp() in quad.rs:259 and quad.rs:534 — the fix for fragility 1 is just aligning the test with existing crate patterns.

~ The ok_or_else + format! pattern is already used at line 173 of the same test file — fix 2 is consistent with existing code.

~ is_finite() is used throughout kornia-algebra, kornia-imgproc, and kornia-3d

All fixes are confined to a single file: crates/kornia-apriltag/tests/data_driven.rs

🤝 Contribution Intent

• I plan to submit a PR to fix this bug
• I'm reporting this bug but not planning to fix it