fix: correct anomaly detection to compare original data vs bounds

[sd2k]

Fixes #67.

Claude's commit description below:

---

• Fixed detectAnomalies(): Now compares predictions vs bounds instead
  actual of vs bounds

• Added null handling: Properly skips out-of-sample predictions (null values)

• Updated type signatures: Support null values in originalValues array

• Extracted testable function: Made detectAnomalies() independently testable

• Added comprehensive test suite with 11 test cases covering:

  • Basic anomaly detection scenarios
  • Extreme outlier cases
  • Null value handling for out-of-sample predictions
  • Edge cases (empty arrays, missing bounds, missing callbacks)
  • Real-world scenarios (gradual drift, sudden spikes, system failures)
  • Integration tests validating the complete fix

Before: Anomaly detection was essentially broken - would never detect
anomalies in real data when models produced good predictions
After: Anomaly detection works correctly, detecting actual anomalies
in original time series data regardless of model prediction quality

📦 Published PR as canary version: 0.5.1--canary.70.15561052406.0

✨ Test out this PR locally via:

npm install [email protected]
npm install [email protected]
npm install [email protected]
npm install @grafana/[email protected]
# or 
yarn add [email protected]
yarn add [email protected]
yarn add [email protected]
yarn add @grafana/[email protected]