#2632 fix(setup) v2: don't require tensor-data alignment for GGUF shards without

说明

`check_shards()` and `whole_shard()` calculated the minimum GGUF shard size as
`data_start + max(tensor end offsets)`. For a shard containing no tensors, this
incorrectly required the file to extend through the aligned tensor-data start,
even though no tensor payload exists.

This caused valid metadata-only GGUF shards to be rejected as truncated. For
example, `Qwen3.8-Flash-Next-UD-Q5_K_XL-00001-of-00006.gguf` was reported as
six bytes short:

- Actual file size: 10,946,618 bytes
- Parsed tensor count: 0
- Aligned data start: 10,946,624 bytes

The downloaded file matched the size and SHA-256 declared in `setup.py`.

## Change

Skip the tensor-payload extent check when a successfully parsed GGUF shard has
no tensors. GGUF parsing, the existing checks for tensor-bearing shards, and
the separate pinned size/SHA-256 verification remain unchanged.

Issue: Resolves https://github.com/Niko1221/Strata/issues/2470 and https://github.com/Niko1221/Strata/issues/2346
#2491