Canvas heatmap: add `normalize` option for row/column percentage normalization

[jklaise]

Problem
There's no way to display a heatmap where cell values are normalised as a percentage of their row or column total in a way that responds to filters. The workaround of pre-computing percentages in a SQL model breaks reactivity - when a user applies a dimension filter, the denominator was computed before the filter, so cells can exceed 100%. Window functions referencing dimension columns are also explicitly disallowed in measure expressions (#4103).

Proposed solution

Add a normalize option to the heatmap canvas component:

- heatmap:
    metrics_view: my_metrics
    normalize: rows        # or: columns, none (default)
    x:
      field: category
      type: nominal
    y:
      field: region
      type: nominal
    color:
      field: event_count
      type: quantitative

When set, Rill fetches raw measure values and computes each cell's percentage of its row (or column) total client-side, after filters are applied. This is purely a chart-layer transform - no changes to the metrics view or model layer needed.

Prior art

stacked_bar_normalized already implements percentage normalization at the chart layer (#6740) - this is the same concept applied to heatmaps. Vega-Lite's joinaggregate transform also handles this natively post-fetch.