fix: add code ops and extract to python api

[shreyashankar]

PR Description

• Closes #461
• Adds Python SDK support for code-based operations and exposes ExtractOp per Python API expectations: code_map, code_reduce, code_filter, and extract now work with docetl.api.Pipeline and typed schemas.
• Enables passing real Python functions to code ops instead of stringified code.

Summary of changes

• SDK schemas
  • Added CodeMapOp, CodeReduceOp, CodeFilterOp, ExtractOp to docetl.schemas.OpType.
• API exports
  • Exported the above ops in docetl.api.__all__.
• Pipeline loader
  • Updated Pipeline._update_from_dict to recognize code_map, code_reduce, code_filter, extract.
• Callable support for code ops
  • docetl.operations.code_operations.*.schema.code now accepts a callable; it’s converted to source with inspect.getsource and bound as transform.
• Tests
  • Added tests in tests/test_api.py for:
    • CodeMapOp with callable transform
    • CodeFilterOp with callable predicate
    • CodeReduceOp with callable group reducer
    • Export presence of ExtractOp
  • Tests are deterministic and avoid network/LLM calls.

Usage example

from docetl.api import Pipeline, Dataset, PipelineStep, PipelineOutput, CodeMapOp

def double(doc: dict) -> dict:
    return {"double": doc["x"] * 2}

pipeline = Pipeline(
    name="example",
    datasets={"input": Dataset(type="file", path="input.json")},
    operations=[CodeMapOp(name="double_x", type="code_map", code=double)],
    steps=[PipelineStep(name="s1", input="input", operations=["double_x"])],
    output=PipelineOutput(type="file", path="out.json"),
    default_model="gpt-4o-mini",
)

Files touched

• docetl/schemas.py: add and export new op schemas; extend OpType
• docetl/api.py: export new ops; handle new op types in Pipeline._update_from_dict
• docetl/operations/code_operations.py: accept callables for code via schema validators
• tests/test_api.py: add tests for code ops and ExtractOp

Backward compatibility

• No breaking changes to existing APIs.
• Code ops now accept either strings or callables; string behavior unchanged.

Docs

• Refreshing docs as well: Python API

Notes

• Callable must be a regular def (no lambda/closures); source is captured via inspect.getsource.
• CodeMap: transform(doc: dict) -> dict
• CodeFilter: transform(doc: dict) -> bool
• CodeReduce: transform(group: list[dict]) -> dict