feat: Implement coerce_types for Timestamp(Second), Interval(MonthDayNano)

[CuteChuanChuan]

Which issue does this PR close?

• Closes Add coerce_types flag to parquet ArrowWriter #1938.

Rationale for this change

Some Arrow types (Timestamp(Second), Interval(MonthDayNano)) have no direct corresponding Parquet logical type. PR #6840 introduced the coerce_types flag and implemented Date64 coercion, but Timestamp and Interval coercion remained unimplemented.

What changes are included in this PR?

Timestamp(Second):

• coerce_types=true: stored as INT64 with LogicalType::Timestamp(MILLIS),
  values multiplied by 1000
• coerce_types=false: stored as raw INT64 without logical type (unchanged)

Interval(MonthDayNano):

• coerce_types=true: stored as 12-byte Parquet INTERVAL, nanoseconds
  truncated to milliseconds (lossy)
• coerce_types=false: stored as 16-byte raw FIXED_LEN_BYTE_ARRAY,
  preserving full nanosecond precision (lossless)

Other:

• Add reader support for MonthDayNano from both 12-byte and 16-byte formats
• Add apply_hint rule for FixedSizeBinary(16) → Interval(MonthDayNano)
• Remove NYI error for writing IntervalMonthDayNanoArray

Are these changes tested?

Yes. 10 new tests added:

• 5 schema-level tests verifying Parquet type output with/without coerce_types
• 2 round-trip tests (write → read) for both Timestamp(Second) and MonthDayNano
• 2 edge-case tests covering overflow boundaries, negative values, nulls, and nanosecond truncation behavior
• 1 regression test confirming YearMonth/DayTime intervals are unaffected by coerce_types

The existing interval_month_day_nano_single_column test was updated from
#[should_panic] to a passing test.

Are there any user-facing changes?

• IntervalMonthDayNanoArray is now supported by the Parquet writer (previously returned a NYI error)
• WriterProperties::set_coerce_types(true) now also affects Timestamp(Second) and Interval(MonthDayNano) columns