feat: Track the table extraction method (#4346)

<!-- CURSOR_SUMMARY -->
> [!NOTE]
> **Low Risk**
> Adds a new optional metadata field and wires it through PDF
partitioning; behavior is additive and low risk aside from potential
downstream consumers expecting a fixed metadata schema.
> 
> **Overview**
> **Adds table-provenance metadata for extracted tables.**
`ElementMetadata` now includes an optional `table_extraction_method`
(e.g., `grid`, `tatr`, `vlm`) and includes it in metadata consolidation.
> 
> During PDF partitioning, the value is propagated from each
`LayoutElement` onto the resulting element metadata, enabling downstream
consumers to identify which table-extraction algorithm produced a given
table. Version is bumped to `0.22.26` and the changelog is updated
accordingly.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
fe6135a. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

Author: vladimir-kivi-ds

CHANGELOG.md

@@ -1,3 +1,9 @@
+## 0.22.26
+
+### Enhancements
+
+- Add `table_extraction_method` field to `ElementMetadata` to track which algorithm produced a table (grid, tatr, vlm). Propagated from `LayoutElement` during PDF partitioning.
+
 ## 0.22.25
 
 ### Enhancements

unstructured/__version__.py

@@ -1 +1 @@
-__version__ = "0.22.25"  # pragma: no cover
+__version__ = "0.22.26"  # pragma: no cover

unstructured/documents/elements.py

@@ -213,6 +213,7 @@ class ElementMetadata:
     text_as_html: Optional[str]
     is_extracted: Optional[str]
     table_as_cells: Optional[dict[str, str | int]]
+    table_extraction_method: Optional[str]  # "grid", "tatr", or "vlm"
 
     # -- used for TableChunk elements to enable table reconstruction --
     table_id: Optional[str]
@@ -267,6 +268,7 @@ def __init__(
         signature: Optional[str] = None,
         subject: Optional[str] = None,
         table_as_cells: Optional[dict[str, str | int]] = None,
+        table_extraction_method: Optional[str] = None,
         table_id: Optional[str] = None,
         chunk_index: Optional[int] = None,
         num_carried_over_header_rows: Optional[int] = None,
@@ -320,6 +322,7 @@ def __init__(
         self.subject = subject
         self.text_as_html = text_as_html
         self.table_as_cells = table_as_cells
+        self.table_extraction_method = table_extraction_method
         self.table_id = table_id
         self.chunk_index = chunk_index
         self.num_carried_over_header_rows = num_carried_over_header_rows
@@ -548,6 +551,7 @@ def field_consolidation_strategies(cls) -> dict[str, ConsolidationStrategy]:
             "subject": cls.FIRST,
             "text_as_html": cls.STRING_CONCATENATE,
             "table_as_cells": cls.FIRST,  # -- only occurs in Table --
+            "table_extraction_method": cls.FIRST,
             "table_id": cls.DROP,  # -- added by chunking, not before --
             "chunk_index": cls.DROP,  # -- added by chunking, not before --
             "num_carried_over_header_rows": cls.DROP,  # -- added by chunking, not before --

unstructured/partition/pdf.py

@@ -1443,6 +1443,9 @@ def document_to_element_list(
                     element.metadata.last_modified = last_modification_date
                 element.metadata.text_as_html = getattr(layout_element, "text_as_html", None)
                 element.metadata.table_as_cells = getattr(layout_element, "table_as_cells", None)
+                element.metadata.table_extraction_method = getattr(
+                    layout_element, "table_extraction_method", None
+                )
 
                 if (isinstance(element, Title) and element.metadata.category_depth is None) and (
                     has_headline