Table data searchability — index table content for retrieval

[iliia-veselov]

What is the problem this feature will solve?

The table counterpart to chart searchability. Tables carry dense, precise figures that are today either flattened into an unusable run of numbers or missed entirely, so a question naming a row, a column or a period will not reliably retrieve the table that answers it.

Sequenced after chart searchability so it can reuse whatever indexing approach that work settles on.

What is the proposed feature or solution?

1. Detect table regions during indexing and preserve their structure — headers, row labels, units, footnotes — rather than flattening them.
2. Index tables so a question naming a row, column or period retrieves the right table.
3. Preserve enough surrounding context that a retrieved table is interpretable on its own.
4. Reuse the approach agreed for charts where it applies, so there is one visual-index path rather than two.

Acceptance criteria

• A question answerable only from a table retrieves that table.
• Retrieved tables retain headers, row labels and units.
• Table hits are not duplicated across chunks.
• The approach is consistent with the chart-searchability implementation.