Add `source` field to extracted variables (filename vs text)

[michaelbeutler]

Allow regex variable extraction to run against the original filename in addition to the extracted document text. Add a source field to ExtractedVariable.

Example Config

variables:
  extracted:
    - name: account_id
      source: filename   # NEW — extract from filename instead of text
      pattern: "[0-9]{8}_.*(?P<account_id>[0-9]{4}\\.[0-9]{4}\\.[0-9]{4})"
    - name: invoice_num
      source: text        # default behavior
      pattern: "INV-(?P<invoice_num>\\d+)"

Implementation

Schema change:

• File: crates/paporg/src/config/schema.rs
• Add source: Option<VariableSource> to ExtractedVariable (line ~65)
• Add enum:

  enum VariableSource { Text, Filename }

  Default: Text

Variable engine change:

• File: crates/paporg/src/config/variables.rs
• Update extract_variables() (line ~36) to accept both text: &str and filename: &str
• For each pattern, check source to decide which string to match against

Pipeline change:

• File: crates/paporg/src/pipeline/runner.rs
• In step_extract_variables() (line ~222): pass the original filename alongside the text

Acceptance Criteria

• source: filename extracts variables from the original filename
• source: text (default) preserves current behavior
• Omitting source defaults to text
• Covered by unit tests for both sources