[python] Align temporal string schema evolution with Java and Native

[JingsongLi]

What changes

Make existing temporal-to-character schema evolution produce Java's text in both Python and Native reads:

• TIME precision follows DateTimeUtils' digit loop, including .0 at whole seconds.
• TIMESTAMP uses a space separator, declared minimum fractional width, and exact nanosecond digits.
• TIMESTAMP_LTZ uses the local time zone without a trailing Z or offset.
• Scalar CHAR/VARCHAR casts apply Java padding/truncation.
• ARRAY/MAP/ROW string rendering uses the same temporal leaf casts and preserves NULL containers/elements.
• Share formatting with existing row-to-string rendering.
• Use integer Gregorian dates and Java's signed millisecond/epoch-day rules rather than restricting Arrow data to Python datetime's year range.

No new Python capability or public API is introduced. These casts and schema changes already existed; this aligns their results with Java and Rust core.

Rust dependency

Depends on apache/paimon-rust#1115. This PR is a draft until that change is merged. The Native CI continues to build paimon-rust from upstream main, so the Rust PR must merge first.

Verification

• 195 REST schema-evolution tests passed with Native planning, reading, writing and committing enabled: 447 reads, 261 writes and commits.
• New tests cover Append, primary-key and Data Evolution tables; nested ARRAY/MAP/ROW values; predicate reads; CHAR/VARCHAR lengths; pre-epoch values, nanoseconds and wide dates.
• LTZ REST tests passed in UTC, America/New_York and Asia/Shanghai, including winter/summer instants.
• 179 related Python tests and 32 subtests passed: existing schema evolution, nested/PK reads, schema manager, row rendering and casting.
• Changed files passed Flake8 and production code passed Python 3.6 grammar validation.
• Re-ran an 85-combination REST cast probe: no remaining differences among the exercised existing Python casts.
• Checked actual Java CastExecutors for temporal formatting, wide years, signed epoch-day narrowing, seconds wrapping and local-zone overflow.

[leaves12138]

Reviewed the temporal formatting and schema-evolution paths, including the focused Python regression tests. I found no blocking issues.