Extract GeoDataFrameExecutor as shared base class for GDF-based executors`

[profsergiocosta]

Summary

Currently, every GeoDataFrame-based executor (LUCCVectorExecutor, future
raster/vector variants) reimplements the same boilerplate for load(),
validate(), and save(). The only genuinely domain-specific method is
run(). This issue proposes extracting a shared GeoDataFrameExecutor base
class to eliminate that duplication.

Motivation

LUCCVectorExecutor.load() and save() are already nearly identical to what
any future GDF executor would need:

• load() — calls load_dataset(), captures checksum, applies column_map
• validate() — checks column_map keys against model spec
• save() — calls save_dataset(), records output_sha256, sets status
• _check_columns() — post-load column presence check

These four concerns are infrastructure, not domain logic. A second executor
(e.g., a raster variant or a new LUCC model) would duplicate them verbatim —
which is the signal that a base class is warranted.

Proposed Design

New file: dissmodel/executor/geodataframe.py

class GeoDataFrameExecutor(ModelExecutor):
    """
    Base class for executors that operate on GeoDataFrames.

    Provides default implementations of load(), validate(), and save()
    that cover the standard GDF contract (load_dataset, column_map,
    checksum, save_dataset). Subclasses only need to implement run().
    """

    output_ext: str = "gpkg"  # subclasses may override (e.g. "geojson")

    def load(self, record: ExperimentRecord) -> gpd.GeoDataFrame: ...
    def validate(self, record: ExperimentRecord) -> None: ...
    def save(self, result: gpd.GeoDataFrame, record: ExperimentRecord) -> ExperimentRecord: ...

    def run(self, data: gpd.GeoDataFrame, record: ExperimentRecord):
        raise NotImplementedError

Reduced executor definition

After this change, LUCCVectorExecutor becomes:

class LUCCVectorExecutor(GeoDataFrameExecutor):
    name = "lucc_vector"

    def run(self, data: gpd.GeoDataFrame, record: ExperimentRecord) -> gpd.GeoDataFrame:
        # only domain logic: Environment, Demand, Potential, Allocation
        ...

Three overridden methods collapse into one.

Scope

In scope:

• Extract GeoDataFrameExecutor into dissmodel/executor/geodataframe.py
• Migrate LUCCVectorExecutor (in dissluc) to inherit from it
• Export GeoDataFrameExecutor from dissmodel.executor
• Update docstrings and type hints

Out of scope (tracked separately):

• Declarative pipeline via TOML (model.pipeline driving component
  instantiation) — more powerful but trades flexibility for convention;
  worth evaluating post-JOSS
• Raster base class (XarrayExecutor or similar) — follow-up issue once
  a second raster executor exists

Acceptance Criteria

• GeoDataFrameExecutor exists in dissmodel/executor/geodataframe.py
• LUCCVectorExecutor inherits from it with only run() overridden
• All existing tests pass without modification
• GeoDataFrameExecutor is exported from dissmodel.executor.__init__
• Docstring explains the output_ext override pattern

Notes

The right moment to implement this is when a second GDF executor is
introduced — the duplication cost becomes concrete and the abstraction boundary
is validated by two real use cases. This issue can be picked up speculatively
before that point if bandwidth allows.