[Feat][Java] Format-neutral physical batch and write contract

Part of #947.

The layering agreed in #947 puts a physical IO boundary between the GraphAr
reader/writer and any concrete file format. Before either vertical can be
built, the project needs one description of a row group that names no format:
a column type, a schema, a row, a batch, and a cursor over batches.

Scope:

- `ColumnType`, `Field`, `Schema` — neutral, recursive type description.
- `Row`, `RecordBatch`, `BatchCursor` — row access with one documented Java
value mapping per column kind.
- `WriteRequest`, `WriteMode`, `PhysicalWriter` — the write half of the
boundary, which shares every one of the types above.

Non-goals for this issue:

- The read request (projection, filter, row range, capability reporting) and
`PhysicalReader`, which follow as their own issue against this base.
- Arrow as the public batch representation.
- Any GraphAr layout, chunking, or query semantics.

The module must not depend on any other GraphAr module.

Start by reading the layering agreed in #947 and the listed contract types: ColumnType, Field, Schema, Row, RecordBatch, BatchCursor, WriteRequest, WriteMode, and PhysicalWriter. Confirm that the physical IO boundary is format-neutral, uses documented Java value mappings, and that the module depends on no other GraphAr module; read-side contracts and Arrow remain out of scope.

Written by the indexing model from the issue text.