stats: styling classes' requires_summary is never checked against the stats the pipeline provides

[paddymul]

Problem

A styling class declares the summary stats it needs in requires_summary, e.g. DefaultMainStyling.requires_summary = ["histogram", "is_numeric", "dtype", "_type"] (buckaroo/customizations/styling.py:67). Nothing checks that the stats pipeline actually provides them. StatPipeline._normalize_inputs treats styling ColAnalysis classes as no-ops (buckaroo/pluggable_analysis_framework/stat_pipeline.py:51), so their requires_summary never enters the DAG validation that @stat functions go through.

When a required stat is missing, what happens depends on how style_column reads it:

• DefaultMainStyling.style_column checks for _type and returns an obj displayer if it is absent (styling.py:77). The column renders plain with no warning.
• If style_column raises instead, StylingAnalysis.style_columns swallows the exception when col_meta is empty and requires_summary is non-empty (buckaroo/dataflow/styling_core.py:470), and falls back to default_styling. It only logs a warning when col_meta is non-empty.

Impact

A styling class used with a backend or stat set that doesn't compute its inputs produces plain columns instead of an error. This matters more now that styling is meant to be backend-generic (#928 types the styling frame argument as DataFrameLike and the analysis lists as ColAnalysis).

Suggested fix

When the pipeline is built, check each styling class's requires_summary against the union of stats the @stat functions provide, and warn (or raise under debug=True) naming the class and the missing keys. Keep the per-column fallback for the startup case where col_meta is legitimately empty.

Context

Identified while typing the dataflow pipeline in #928.