Refactor scalar min/max dispatch into function-based helpers (#22062)

## Which issue does this PR close?

* Closes #22061.

## Rationale for this change

The existing `min_max_scalar_impl!` macro combined dispatch, validation,
recursion, and control flow into a large macro expansion, making the
logic difficult to debug, review, and test in isolation.

This change moves scalar min/max dispatch to a function-based
implementation while preserving existing behavior, error messages, and
dictionary handling semantics.

## What changes are included in this PR?

* Replaced the `min_max_scalar_impl!` macro with function-based dispatch
logic.
* Added helper functions to separate responsibilities:

  * `min_max_option`
  * `min_max_float_option`
  * `ensure_decimal_compatibility`
  * `min_max_generic_scalar`
  * `min_max_interval_scalar`
  * `min_max_dictionary_scalar`
  * `min_max_scalar_same_variant`
* Kept `min_max_scalar` as the main entry point and delegated behavior
through helper functions.
* Preserved:

  * existing min/max behavior across scalar types
  * dictionary comparison and rewrapping semantics
  * decimal precision/scale validation behavior
  * float `NaN` handling via `total_cmp`
  * existing error classifications and messages
  * compacting behavior for generic scalar comparisons
* Simplified macro usage by retaining only lightweight
ordering-selection macros.

## Are these changes tested?

Yes.

Added unit tests covering:

* core scalar min/max parity behavior
* float `NaN` handling using `total_cmp`
* decimal mismatch error preservation
* fixed-size binary mismatch error preservation
* mixed interval comparison error preservation

Existing dictionary comparison tests are also preserved.

## Are there any user-facing changes?

No. This PR is intended to be a refactor only and preserves existing
externally visible behavior.

## LLM-generated code disclosure

This PR includes LLM-generated code and comments. All LLM-generated
content has been manually reviewed and tested.

Author: kosiew