feat(predictions): highlight anomalies on Detect pipeline detail page

[roman1887]

Context

The Detect (Anomaly Detection) pipeline detail page (/predictive-intelligence → click a Detect pipeline) shows a Predictions table with is_anomaly and anomaly_score columns, but anomalous rows are visually indistinguishable from normal ones — just a 1 vs 0 in a column.

For a user skimming the table, this defeats the entire purpose of anomaly detection: viewers can't spot the outliers at a glance.

Expected behavior

On the Detect detail page Predictions table:

• Rows where is_anomaly = 1 should render with a red row background (subtle red tint) OR red text on is_anomaly and anomaly_score cells
• The anomaly_score column should additionally use a color gradient — higher scores → more intense color
• Sort the table by anomaly_score DESC by default so the most anomalous rows surface first
• A small legend/key above the table explaining the color coding

Acceptance criteria

• Anomalous rows are visually obvious without reading the is_anomaly column
• Default sort surfaces highest-scoring anomalies first
• Works on the existing demo data (Avg metric per project Anomaly Detection)
• Uses existing Material UI / Tailwind theme — no new color libraries
• Tested manually on demo.actyze.ai

Motivation

• Customer demos: "the model flagged unusual rows" is a much harder sell with a uniform-looking table
• Marketing video script Adds a new query page view #4 (Anomaly Detection 60s) is blocked on this — Scene 2 explicitly needs a clearly anomalous row when zoomed in

Once shipped + deployed to demo.actyze.ai, the Anomaly Detection video can be recorded.