fix(metrics): make convert_types handle pd.NA / pd.NaT (#1844)

[jbbqqf]

Closes #1844 (and unblocks #1616 — same root cause).

Why

convert_types is called for every label key flowing into
ByLabelCountValue.counts / .shares. On main it ends with:

if val is None or np.isnan(val):
    return val
raise ValueError(f"type {type(val)} not supported as Label")

np.isnan(pd.NA) raises TypeError: boolean value of NA is ambiguous
and np.isnan(pd.NaT) raises TypeError: ufunc 'isnan' not supported for the input types .... Any preset (e.g. DataSummaryPreset,
DataDriftPreset) that touches a column with a nullable pandas dtype
("string", "Int64", …) and any pd.NA value blows up before the
result can be built, surfacing as a confusing
pydantic.ValidationError: boolean value of NA is ambiguous.

Fix

After the existing bool / int / str fast paths, guard the NA branch with
pd.isna(val) (the only NA-detection helper that handles all of
None / np.nan / pd.NaT / pd.NA). Pandas-flavored NA collapses to
None; numpy.nan is preserved as a float so the existing serializer
keeps mapping it to the string "nan" — see the preserved
test_by_label_count_value test.

Reproduce BEFORE/AFTER yourself (copy-paste)

# from a fresh checkout
git clone https://github.com/evidentlyai/evidently.git && cd evidently
uv venv .venv && . .venv/bin/activate
uv pip install -e . pytest pytest-timeout pytest-asyncio

# === BEFORE (origin/main) — expected: 3 failures ===
git checkout origin/main
git fetch https://github.com/jbbqqf/evidently.git fix/1844-convert-types-pd-na
git checkout FETCH_HEAD -- tests/future/test_metric_types.py
pytest tests/future/test_metric_types.py::test_convert_types_handles_pandas_na tests/future/test_metric_types.py::test_by_label_count_value_handles_pd_na_key -v
# Expected: 3 failed — TypeError: boolean value of NA is ambiguous

# === AFTER (this PR) — expected: all green ===
git checkout FETCH_HEAD -- src/evidently/core/metric_types.py
pytest tests/future/test_metric_types.py -v
# Expected: 5 passed

What I ran locally

$ pytest tests/future/test_metric_types.py -v
tests/future/test_metric_types.py::test_by_label_count_value[input0-output0]         PASSED
tests/future/test_metric_types.py::test_convert_types_handles_pandas_na[value0]      PASSED
tests/future/test_metric_types.py::test_convert_types_handles_pandas_na[value1]      PASSED
tests/future/test_metric_types.py::test_convert_types_preserves_existing_contracts   PASSED
tests/future/test_metric_types.py::test_by_label_count_value_handles_pd_na_key       PASSED
=========================== 5 passed, 4 warnings in 0.03s ============================

$ pytest tests/future/metrics tests/future/test_metric_types.py tests/future/presets -q
1458 passed, 1363 warnings in 48.53s

Failing assertion before the fix (copied verbatim from pytest -v on
origin/main with the regression test applied):

E   pydantic.v1.error_wrappers.ValidationError: 2 validation errors for ByLabelCountValue
E   counts
E     boolean value of NA is ambiguous (type=type_error)
E   shares
E     boolean value of NA is ambiguous (type=type_error)

Edge cases covered by the regression tests

Input	Old behavior	New behavior
pd.NA	TypeError: boolean value of NA is ambiguous	returns None
pd.NaT	TypeError: ufunc 'isnan' not supported	returns None
np.nan	returns nan (serializer → "nan")	unchanged
None	returns None	unchanged
True, 42, "class_a"	returned untouched	unchanged
ByLabelCountValue with pd.NA key	crashes in __init__	builds + serializes fine

---

Disclosure: I drafted this PR with help from Claude Code while triaging
older issues. The reproduction, fix, and test runs above were executed
locally; outputs are copied verbatim.