feat(arrow): add opt-in mixed type coercion - #1591

Closes #867.

Summary

• Add auto_coerce=False to plot(), upload(), and to_arrow() so Arrow conversion remains strict by default.
• When auto_coerce=True, retry failed pandas/cuDF Arrow conversion by probing object columns one at a time and coercing only Arrow-incompatible columns to strings.
• Emit an info log naming coerced columns instead of warning or silently coercing.
• Update docs, changelog, Plottable protocol, and anchored regression tests.

Coercion heuristic

Graphistry first runs the normal Arrow conversion. If pandas/cuDF conversion fails with ArrowTypeError/ArrowInvalid and auto_coerce=True, it tests object columns individually with the same Arrow conversion path. Only columns that fail individual Arrow conversion are coerced to strings; Arrow-compatible object columns are left alone. Pandas coercion uses nullable string dtype so missing values remain Arrow nulls. cuDF uses its string dtype path. This keeps the fallback narrow and mirrors the user workaround from #867 while preserving strict default behavior.

Info-log sample:

Auto-coerced mixed-type columns to string for Arrow conversion: ['amount']

LOC buckets

• Production: +61 / -60, net +1 LOC
• Tests / fixtures / baselines: +129 / -64, net +65 LOC
• Comments / docs / changelog: +17 / -10, net +7 LOC

Compiler-plan surface touched: no. This is public Arrow-conversion API surface only; no GFQL compiler-plan routes, dispatch contracts, error-code/defer-code context, source spans, pass/verifier hooks, schema/type-system hooks, remote schema hooks, or compat executor paths were touched.

Validation

• python3 -m pytest -q graphistry/tests/test_plotter.py::TestPlotterArrowConversions -> 25 passed, 7 skipped, 1 xfailed
• python3 -m pytest -q graphistry/tests/test_plotter.py graphistry/tests/test_validate_settings.py graphistry/tests/test_schema_artifacts.py graphistry/tests/test_viz_settings.py -> 83 passed, 10 skipped, 4 xfailed
• python3 bin/changed_line_coverage.py --data-file /tmp/pr1591.coverage --base-ref origin/master --head-ref HEAD --min-percent 80 --output-dir /tmp/pr1591-changed-coverage -> passed, 19/23 = 82.61%
• python3 -m graphistry.devschemas.export --check
• ./bin/ruff.sh graphistry/PlotterBase.py graphistry/Plottable.py graphistry/exceptions.py graphistry/tests/test_plotter.py
• ./bin/typecheck.sh graphistry/PlotterBase.py graphistry/Plottable.py graphistry/exceptions.py
• git diff --check
• ssh dgx-spark 'cd repos/pygraphistry-issue-867-dgx && RAPIDS_VERSION=25.02 PROFILE=basic WITH_GPU=1 WITH_IMAGE_BUILD=0 TEST_FILES="graphistry/tests/test_plotter.py::TestPlotterArrowConversions::test_table_to_arrow_cudf_auto_coerce_smoke graphistry/tests/test_plotter.py::TestPlotterArrowConversions::test_api3_cudf_to_arrow_memoization graphistry/tests/test_plotter.py::TestPlotterArrowConversions::test_api3_cudf_to_arrow_memoization_forgets" docker/test-rapids-official-local.sh' -> cudf 25.02.02, cuml 25.02.01, Python 3.12.9, 3 passed
• ssh dgx-spark 'cd repos/pygraphistry-issue-867-dgx && RAPIDS_VERSION=26.02 PROFILE=basic WITH_GPU=1 WITH_IMAGE_BUILD=0 TEST_FILES="graphistry/tests/test_plotter.py::TestPlotterArrowConversions::test_table_to_arrow_cudf_auto_coerce_smoke graphistry/tests/test_plotter.py::TestPlotterArrowConversions::test_api3_cudf_to_arrow_memoization graphistry/tests/test_plotter.py::TestPlotterArrowConversions::test_api3_cudf_to_arrow_memoization_forgets" docker/test-rapids-official-local.sh' -> cudf 26.02.01, cuml 26.02.00, Python 3.13.12, 3 passed
• Review skill converged: W1 fixed nullable-string coercion; W2/W3 clean; W4 clean after cuDF smoke addition; W5 clean after CI test mock fix; W6 clean after changed-line coverage test amplification

Typecheck note: ./bin/typecheck.sh ... graphistry/tests/test_plotter.py still hits existing test typing gaps (pytest stubs missing and unrelated COMPLEX_EMPTY annotation). Production touched files typecheck cleanly.