[SPARK-58766][PYTHON][TEST][FOLLOWUP] Render NaT-sentinel time64 scalars in golden files

[Spenserrrr]

What changes were proposed in this pull request?

Add a ValueError branch to _scalar_str in python/pyspark/testing/goldenutils.py so a time64[ns] scalar holding INT64_MIN renders as the marker NaT collision, beside the existing temporal overflow and raw-bytes fallbacks, and regenerate the four affected from_pandas type-scalar golden cells (int64:standard and Int64:standard x time64[ns], safe and unsafe).

Rendering a time64[ns] scalar goes through pandas, because Python's datetime.time has only microsecond resolution. pandas reserves INT64_MIN as its NaT sentinel, so it refuses that one value with ValueError: NaTType does not support time. The branch is gated on both the type and the value, so any other ValueError still propagates.

Why are the changes needed?

PyArrowArrayFromPandasTypeScalarTests currently fails on master. The four cells above read ERR@ValueError, which by convention means the conversion raised — but the conversion succeeds and only the rendering failed. The broad try/except in the cell helper recorded the rendering failure as if it were a conversion error, so narrowing that try/except in #58091 turned the mislabelling into a test error.

Does this PR introduce any user-facing change?

No. Test-only.

How was this patch tested?

python/pyspark/tests/upstream/pyarrow/ passes (63 passed, 1 skipped), and a sweep over pyarrow 18-25 x pandas 2/3 is 16/16 with no overrides entries.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Claude Opus 5)

[Spenserrrr]

Hi @zhengruifeng @Yicong-Huang! Could you take a look at this when you have time? It fixes PyArrowArrayFromPandasTypeScalarTests, which currently fails on master, so it may surface on other PRs' CI too. My #58091 exposed it, and I wrote the reason in the PR description. CI here is green, including the pyspark-core shard that's the one failing on master. Thanks!

[zhengruifeng]

0 blocking, 0 non-blocking, 0 nits.
No findings; the fallback is narrowly scoped and the golden updates match the exercised matrix cells.

Verification

I traced the from_pandas test helper and confirmed ERR cells are produced only when pa.Array.from_pandas raises, before result rendering. I also checked that the type-scalar matrix has one time64[ns] target and the changed int64/Int64 source rows, then locally evaluated the renderer for pa.time64('ns') with pandas' NaT sentinel; it produces the regenerated NaT collision cell.

[zhengruifeng]

Merge Summary:

• merged into master 08e97de
• merged into branch-4.x 27e2411

Posted by merge_spark_pr.py