[Variant] Align cast logic between `variant_get` and `cast` kernel for temporal types (Timestamp/Date/Time)

**Is your feature request related to a problem or challenge? Please describe what you are trying to do.**

This is part of #8982 which wants to support the temporal types (`Timestamp*` incl. NTZ + timezone, `Date32`/`Date64`, `Time32*`/`Time64*`).

Currently the temporal cast logic in `parquet-variant-compute` (`type_conversion.rs`) hand-rolls conversion/truncation logic that *mimics* `arrow-cast` rather than reusing it. For example, several `Time*`/`Timestamp*` impls return `None` when there is leftover sub-second precision instead of truncating the way `arrow-cast` does (`lossy` is the default in the cast kernel).

**Describe the solution you'd like**

Align the logic of casting from/to the temporal types in `parquet-variant` with `arrow-cast` by reusing `arrow-cast`'s scalar conversion primitives (the same approach #9688 took for decimals, where `rescale_decimal` / `parse_string_to_decimal_native` / `single_float_to_decimal` were exposed and reused) instead of re-implementing the logic. This keeps a single source of truth for the cast semantics and lets the perfect-shredding fast path (`can_use_perfect_shredding_arrow_cast`, added in #9862) be widened to cover the truncating temporal pairs it currently excludes.

**Describe alternatives you've considered**

**Additional context**

cc @klion26 @AdamGS

Start in parquet-variant-compute/type_conversion.rs and compare its temporal conversions with the scalar conversion primitives in arrow-cast. Review can_use_perfect_shredding_arrow_cast and the temporal pairs currently excluded; done means the conversions reuse arrow-cast semantics, including lossy sub-second truncation, and the perfect-shredding path covers the supported pairs.

Written by the indexing model from the issue text.