[Repo Assist] feat: add explicit-thresholds overloads to multiLabelThresholdMap and calculateMultiLabelROC

[github-actions]

🤖 This is an automated pull request from Repo Assist, an AI assistant.

Closes #198 (partial — the explicit-thresholds items).

Summary

Adds overloads that accept a caller-supplied thresholds: float [] to:

• ComparisonMetrics.multiLabelThresholdMap
• ComparisonMetrics.calculateMultiLabelROC

Previously the only option was the zero-argument form that computes BCMs for every distinct prediction value across all labels. For large datasets this is expensive — you may have tens of thousands of distinct prediction scores but only need a coarse 10-point ROC grid.

// fast preview with 5 thresholds instead of all distinct values
let roc =
    ComparisonMetrics.calculateMultiLabelROC(
        actual      = actualLabels,
        predictions = perLabelPredictions,
        thresholds  = [|0.9; 0.7; 0.5; 0.3; 0.1|]
    )
```

## Root cause / design

The no-argument form collects `Array.concat (Array.map snd predictions) |> Array.distinct |> Array.sortDescending` and passes the result to `BinaryConfusionMatrix.thresholdMap`.  When the caller provides their own threshold list they can skip that O(n) collection step and evaluate only the points they care about.

The existing no-argument overloads are **refactored to delegate** to the new ones, eliminating the previously duplicated internal logic.

## Trade-offs

- **Caller responsibility**: when an explicit threshold list is provided the user must ensure the values are sensible (sorted descending is not required — `BinaryConfusionMatrix.thresholdMap` handles any order, but descending is conventional for ROC curves).
- **Prefix threshold**: the output still includes one entry above the global max (`max(all predictions) + 1`) so the first point on the ROC curve always represents the all-negatives classifier.

## Additional fixes in this PR

- Fixed unmatched parenthesis in the original `calculateMultiLabelROC` body (no behaviour change, just compilation cleanliness).
- Renamed unused `k` binding in lambda to `_k`.

## Test Status

5 new test cases added to `Testing.fs`:

```
multi-label threshold map with explicit thresholds
  ✓ explicit thresholds: result length
  ✓ A: explicit threshold 0-9
  ✓ B: explicit threshold 0-5
  ✓ C: explicit threshold 0-1
  ✓ micro-average present
  ✓ macro-average present

Full suite: 1199/1199 passed on .NET 8.

Generated by 🌈 Repo Assist, see workflow run. Learn more.

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@96b9d4c39aa22359c0b38265927eadb31dcf4e2a