Unify plain/gz PDB parsing and fix short-record handling in gz reader

[nz-mtra]

What this PR changes

This PR removes divergence between plain and gz PDB parsing by introducing one shared line parser in pdb.rs. Fixes #45 .

Changes included:

1. Added shared helper parse_pdb_lines to centralize:
   • short-line guard before record prefix matching
   • MODEL counting
   • first-model-only behavior
   • ATOM parsing/update flow
2. Updated:
   • read_structure to use shared parser
   • read_structure_from_gz to decode and then use the same parser
3. Replaced gz decode unwrap with map_err to avoid panic on decode/read errors.
4. Added regression tests in pdb.rs:
   • test_read_pdb_gz_with_end_record (END-terminated content parity plain vs gz)
   • test_plain_and_gz_model_handling_match (multi-model handling parity plain vs gz)
5. Removed now-unused top-level Write import warning in pdb.rs.

Why

Previously, gz and plain PDB readers used separate parsing paths. That made behavior drift possible and exposed gz parsing to short-record edge cases. The new shared parser keeps behavior aligned and prevents this class of divergence moving forward.

Validation

Executed and passing:

1. cargo test test_read_pdb_gz_with_end_record -- --nocapture
2. cargo test test_plain_and_gz_model_handling_match -- --nocapture
3. cargo test --workspace --all-targets --release

Notes

The new regression tests assert plain/gz behavior parity and non-crashing behavior for END-terminated inputs, which directly targets the reported bug class.