Give user more control over docling processing

[ppinchuk]

Also adds option to use fast ELM parsing as fallback if docling processing fails.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

Suppressed comments (3)

examples/execution_basics/README.rst:151

• The JSON example is invalid (missing a comma after true and includes a trailing comma), and it also uses docling_timeout instead of the implemented document_timeout key.

        "pdf_pipeline_options": {
            "do_ocr": true,
            "do_table_structure": true
            "docling_timeout": 1800,
        }

compass/services/cpu.py:592

• _validate_docling_timeout validates the value taken from pdf_pipeline_options["document_timeout"], but the error message refers to docling_timeout, which is confusing for users configuring Docling.

def _validate_docling_timeout(timeout):
    """Validate the configured Docling deadline"""
    if (
        isinstance(timeout, bool)
        or not isinstance(timeout, (int, float))
        or not math.isfinite(timeout)
        or timeout <= 0
    ):
        msg = "`docling_timeout` must be a positive number of seconds"
        raise COMPASSValueError(msg)

examples/execution_basics/README.rst:140

• The docs describe configuring docling_timeout, but the implementation and tests use Docling’s document_timeout option (passed via pdf_pipeline_options). As written, this will mislead users into setting a key that the code doesn’t read.

This issue also appears on line 147 of the same file.

**Docling conversion deadline**
When using the Docling file loader, set ``docling_timeout`` in
``file_loader_kwargs.pdf_pipeline_options`` to apply a wall-clock deadline to each document.
COMPASS runs the conversion in a disposable child process and terminates that