initialize_pipeline should have extract_code and extract_formulas default value as False

[yqliving]

Bug

When I call initialize_pipeline (InputFormat.PDF) first, the pipeline is cached, then when I call the convert function, it should get the pipeline from the cache directly, but now it does reinitialize. I found the reason is:
When initialize_pipeline() is called, the cache hash is computed with code_formula_options.extract_code=True, extract_formulas=True (defaults)
however, during pipeline initialization, _init_models() mutates the options:

code_formula_opts.extract_code = self.pipeline_options.do_code_enrichment
code_formula_opts.extract_formulas = self.pipeline_options.do_formula_enrichment

So it makes initialize pipeline twice.
So initialize_pipeline should have extract_code and extract_formulas default value as False.

Steps to reproduce

1. Define docling converter
2. call converter.initialize_pipeline(InputFormat.PDF) first
3. then call converter.convert()

Docling version

docling==2.73.1
...

Python version

3.11

[majiayu000]

Hi, I've investigated this and confirmed the root cause: _init_models() in StandardPdfPipeline mutates pipeline_options.code_formula_options in-place (setting extract_code and extract_formulas), which changes the hash computed by _get_pipeline_options_hash(). This causes a cache miss when convert() calls _get_pipeline() after initialize_pipeline().

Fix approach: Use model_copy(update=...) instead of mutating the shared pipeline_options object. This creates a local copy with the updated values while keeping the original options (and their hash) stable.

I have a PR ready with the fix and a regression test. Will open it shortly.