feat/improved bias calculation

[Mark-0909]

Major changes:

• updated weights based on the discussed websites
• force database session to expire old searches to eliminate state leakage.
• stream can now track and dedup processed ids.

BIAS IMPLEMENTATION

A. Bias Divergence (The "Echo Chamber" Detector)
Goal: Measures how diverse your source set is.

The Logic: It looks at the spread of news sources across the political spectrum (from Left to Right).
The Math: It calculates the Standard Deviation of all source bias values. If all sources are in the same location (e.g., all are "Left-Center"), the deviation is 0. If they are spread from far-left to far-right, the deviation is high.
The Scale:
1.0 (Maximum Divergence): You have a perfectly represented sample of the whole political spectrum.
-1.0 (Minimum Divergence): You are in an echo chamber; all sources share the same ideological lean.

B. Bias Consistency (The "Polarization" Detector)
Goal: Detects if sources are "picking sides" based on their politics.

The Logic: It calculates the correlation between a source's lean and the verdict they gave.
The Math: It uses the Absolute Pearson Correlation Coefficient (r).
If Left sources always say "Support" and Right sources always say "Refute," then r is 1.0.
If their verdicts have nothing to do with their bias (e.g., everyone agrees or the disagreement is random), r is 0.0.
The Scale:
1.0 (High Consistency/Polarization): The claim is highly partisan. Ideology is a perfect predictor of the verdict.
-1.0 (No Consistency): There is no partisan pattern. The media is either in consensus or the disagreement is not driven by bias.

[github-actions]

## Accuracy Check Results

- Normal dataset (jomTestCases): **71.72%** (71/99)
- Negated dataset (negatedClaims): **81.82%** (81/99)

- Evaluation step outcome: success/success
- Run: https://github.com/jomkv/truescope-api/actions/runs/24721485779