[SPARK-52904][PYTHON] Enable convertToArrowArraySafely by default

[benrobby]

What changes were proposed in this pull request?

• this enables spark.sql.execution.pandas.convertToArrowArraySafely by default
• I am also adjusting unit tests that previously relied on implicit conversions (truncating nanosecond timestamps to microseconds / loss of precision, int overflows) and now start to fail with the new default.

Why are the changes needed?

• This change aligns pyspark UDF behavior to ANSI SQL behavior in the rest of Spark. On integer overflow, the standard behavior is to throw an error. Users can and should handle such overflow or truncation cases explicitly.

Does this PR introduce any user-facing change?

• Yes. This throws errors on int overflow, float truncation, and loss of precision when truncating timestamps. Citing PySpark's upgrade docs:

    =======================================  ================  =========================
    PyArrow version                          Integer overflow  Floating point truncation
    =======================================  ================  =========================
    0.11.0 and below                         Raise error       Silently allows
    > 0.11.0, arrowSafeTypeConversion=false  Silent overflow (returns 0)   Silently allows
    > 0.11.0, arrowSafeTypeConversion=true   Raise error       Raise error
    =======================================  ================  =========================

How was this patch tested?

• adjusted unit tests

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

No

[benrobby]

the diff above looks unfortunate, this is merely indented by one more level.

[benrobby]

@HyukjinKwon @zhengruifeng @asl3 could you take a look?

[asl3]

should we add a note to our migration guide docs for the user-facing change?