feat(predictions): add forecast line chart to pipeline detail page

[roman1887]

Context

The Forecast pipeline detail page (/predictive-intelligence → click a Forecast pipeline) currently shows:

• Status / Last Trained / Output Table
• Accuracy metrics (MAPE / MAE / RMSE / R²)
• Configuration (target column, horizon, model)
• A raw predictions table

It does not show a chart of the forecast. Anyone evaluating the product expects a chart — historical actuals next to the predicted line, with confidence band — and right now we hand them a numeric table instead.

Expected behavior

On the Forecast pipeline detail page, render a line chart above the Predictions table (between Configuration and the table):

• X-axis: time (the original KPI's timestamp column + predicted_at continuation)
• Y-axis: the target column (e.g. daily_events, revenue)
• Two series:
  • Historical actuals — gray bars or muted line for the trailing window of KPI data feeding the pipeline (e.g. last 60 days)
  • Predicted values — Actyze blue line (#5d6ad3) extending past the historical range for the forecast horizon
• Optional (nice-to-have, not blocking): confidence band shaded around the predicted line if the model emits one
• Hover tooltips with exact values

Chart should fill the card width. Aspect roughly 16:6 so it sits visually between the two existing sections.

Acceptance criteria

• Chart renders on the Forecast pipeline detail page only (Classify / Estimate / Detect get their own visualizations in follow-up issues)
• Pulls historical KPI data from the source KPI's materialized table (kpi_data.kpi_<slug>) for at least the past 2 × horizon days
• Pulls predictions from the pipeline's output table (prediction_data.pred_<slug>_<horizon>)
• Empty / loading / error states handled (no infinite spinner, no blank page)
• Works on the existing demo data (Daily Engagement Trend Forecast - 30d)
• Tested manually on demo.actyze.ai before close

Technical notes

• Plotly is already in the frontend stack — use it. No new charting deps.
• Backend probably needs one endpoint that returns historical + predicted rows aligned by timestamp; the frontend should not stitch two queries together.
• The Forecast pipeline's predicted_at column today is a creation timestamp, not the forecast date — the backend response needs to expose the actual forecast date axis (KPI's time column + horizon offsets).

Motivation

• Customer demos: viewers expect a chart, not a table. This is the single biggest UX gap I hit while doing recon for the LinkedIn video series.
• Marketing video script Improve dashboard layout and fix scrolling issues #1 (Forecast 60s) is blocked on this — Scene 3 explicitly needs the chart visual.

Once shipped, deploy to demo.actyze.ai, then we can record the Forecast video and close this.