Unify granularity / field-name parsing across temporal functions #22033

Is your feature request related to a problem or challenge?

Working on #18319 I noticed that date_trunc defines its own DateTruncGranularity enum.

date_part solves the same parsing problem differently — it delegates to arrow_cast::parse::IntervalUnit::from_str:

This leads to a user-visible inconsistency between the two functions. For example, date_part accepts "mon" | "mons" | "month" | "months", while date_trunc only accepts "month".

IntervalUnit, however, is the wrong enum for either of these functions — it exists to parse Postgres INTERVAL literals (INTERVAL '1 year 2 months'), which is why it has Century / Decade but no Quarter (Postgres rejects INTERVAL '1 QUARTER'). The right counterpart is arrow_arith::temporal::DatePart, which represents the date-part field namespace, includes Quarter, and is #[non_exhaustive]:
https://github.com/apache/arrow-rs/blob/8091f3f17b2de355f7c47e7a0907000d308f8f3e/arrow-arith/src/temporal.rs#L38-L79

Describe the solution you'd like

I propose:

• Add a FromStr impl for arrow_arith::DatePart (with the Postgres alias map currently inside IntervalUnit::from_str plus the date-part-only field names).
• Migrate date_part, date_trunc, and the Spark variants (date_trunc, trunc, time_trunc, date_part) to the new parser. The unparser sites in sql/src/unparser/utils.rs can also use it as an input canonicalizer.

Describe alternatives you've considered

No response

Additional context

No response