Add ZINC molecular graph dataset

[Uneeb808]

This PR adds the ZINC molecular graph dataset.

~250,000 molecular graphs with a penalized logP regression target (y = logP - SAS - cycles). Includes the standard 12k benchmark subset from "Benchmarking Graph Neural Networks" (Dwivedi et al. 2020).

Features:

• Full (~250k graphs) and subset (12k) variants
• train/val/test splits
• Node features: atom type (28 classes)
• Edge features: bond type (4 classes)
• Graph-level regression target (Float32)

Note: Raw data is distributed as PyTorch pickle files which Pickle.jl cannot deserialize directly. I preprocessed these to flat NPZ files and hosted them on my fork's releases. Happy to move to JuliaML releases on merge.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 80.95238% with 12 lines in your changes missing coverage. Please review.
✅ Project coverage is 47.75%. Comparing base (3d40a68) to head (ae14f0e).
⚠️ Report is 14 commits behind head on master.

Files with missing lines	Patch %	Lines
src/datasets/graphs/zinc.jl	80.64%	12 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #250      +/-   ##
==========================================
+ Coverage   45.94%   47.75%   +1.80%     
==========================================
  Files          50       54       +4     
  Lines        2501     2689     +188     
==========================================
+ Hits         1149     1284     +135     
- Misses       1352     1405      +53     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Uneeb808]

This PR addresses #239