[FEATURE][Java] Add chunk and edge-range value primitives

### Describe the enhancement requested

Add the small, storage-free value primitives needed to translate GraphAr ordered-adjacency offsets into exact half-open edge-chunk ranges. This is the first core slice of the pure-Java ordered adjacency reader; metadata URI resolution and physical reading follow separately.

Component(s): Java

## Scope

- Add `graphar-core` with long-safe chunk-index, in-chunk offset, and chunk-count arithmetic.
- Add validated half-open `ChunkRange` and `EdgeRange` values, including correct end-exclusive edge-chunk selection.
- Add unit tests for empty ranges, chunk boundaries, and invalid IDs/sizes/offsets.

## Non-goals

No storage, Parquet, S3, Arrow, metadata URI resolution, GraphAr reader facade, offsets file decoder, or adjacency data read.

## Compatibility and acceptance

The API remains Java 11 compatible and depends on no physical IO module. It must reject negative IDs and non-positive chunk sizes. A range `[begin,end)` selects no chunks when empty; nonempty ranges include the chunk containing `end - 1`, not an extra chunk at `end`. Focused Maven verification and pre-commit must pass.

## Links

Relates to #947, #944, and architecture umbrella #756. The follow-up metadata resolver will use `EdgeInfo` URI methods rather than reconstruct GraphAr paths.

Start in the new graphar-core module and review the requested chunk-index, in-chunk offset, chunk-count, ChunkRange, and EdgeRange primitives. Add the unit tests described for empty ranges, chunk boundaries, and invalid IDs, sizes, and offsets, then run focused Maven verification and pre-commit; done means Java 11 compatibility and all stated range validations pass.

Written by the indexing model from the issue text.