[Parquet] Support per-column encodings in parquet-rewrite

### Is your feature request related to a problem or challenge?

`parquet-rewrite --encoding` currently applies one encoding to every column. This makes type-specific encodings difficult to use with mixed schemas.

For example, ALP supports only `FLOAT` and `DOUBLE`. Applying it globally to a file containing other physical types fails after output-file creation. Other type-specific encodings have similar behavior.

This depends on #9372 and should be worked on after that PR is merged.

### Describe the solution you would like

Add per-column encoding selection to `parquet-rewrite`, while retaining the global encoding option as a default.

Validate the selected encoding against each column physical type before rewriting begins. Unsupported combinations should return an actionable error, for example that ALP supports only `FLOAT` and `DOUBLE`, without leaving an incomplete output file.

Start by reviewing the parquet-rewrite implementation and the changes from dependent issue #9372. Trace how --encoding is parsed and applied across columns, then define the per-column selection and pre-rewrite physical-type validation. Done means mixed-schema files can use column-specific encodings, the global option remains the default, and unsupported combinations fail before an incomplete output file is created.

Written by the indexing model from the issue text.