docs: fix QQA/CPRA paper citations, adopt TMLR title rename, broaden PyPI classifiers

The QQA4CO repo had three places where the QQA paper (Ichikawa & Arai,
ICLR 2025) and the CPRA paper (Ichikawa & Iwashita, TMLR 2025) were
silently swapped or labelled with their obsolete arXiv preprint titles:

- src/qqa/__init__.py docstring referenced the CTRA preprint title
  ("Continuous Tensor Relaxation ...") and attributed it to Ichikawa & Arai
  at ICLR 2025. The actual ICLR 2025 paper by those authors is the QQA
  paper, "Optimization by Parallel Quasi-Quantum Annealing with
  Gradient-Based Sampling" (OpenReview 9EfBeXaXf0, arXiv:2409.02135).
- notebooks/cra_pignn_example.ipynb had the same swap in its "QQA paper"
  bullet. Replaced with the real QQA citation and added a separate
  CPRA citation (Ichikawa & Iwashita, TMLR 2025).
- notebooks/cpra_pignn_example.ipynb still used the algorithm name
  ("Continual Parallel Relaxation Annealing for Diverse Solutions in
  Combinatorial Optimization") as the paper title in the intro link and
  bibtex. The actual TMLR title is "Continuous Parallel Relaxation for
  Finding Diverse Solutions in Combinatorial Optimization Problems"
  (the paper was renamed when accepted at TMLR; the older "Continuous
  Tensor Relaxation" arXiv preprint title is obsolete).

Also broadens pyproject.toml PyPI classifiers (Environment :: Console,
Environment :: GPU :: NVIDIA CUDA, Intended Audience :: Education /
Developers, OS-specific tags, Topic :: Mathematics / Physics, Typing ::
Typed) to improve PyPI discoverability for the awesome-list / PWC
campaign.

Author: Yuma-Ichikawa

notebooks/cpra_pignn_example.ipynb

@@ -7,7 +7,7 @@
    "source": [
     "# CPRA walkthrough — diverse solutions in a single GNN run\n",
     "\n",
-    "This notebook is a minimal tour of **CPRA** (*Continual Parallel Relaxation Annealing*), the multi-replica extension of CRA-PI-GNN published in *Transactions on Machine Learning Research* (2025): [\"Continual Parallel Relaxation Annealing for Diverse Solutions in Combinatorial Optimization\"](https://openreview.net/forum?id=ix33zd5zCw).\n",
+    "This notebook is a minimal tour of **CPRA** (*Continual Parallel Relaxation Annealing*), the multi-replica extension of CRA-PI-GNN published in *Transactions on Machine Learning Research* (2025): [\"Continuous Parallel Relaxation for Finding Diverse Solutions in Combinatorial Optimization Problems\"](https://openreview.net/forum?id=ix33zd5zCw).\n",
     "\n",
     "CPRA trains a **single shared GCN backbone** that produces ``R`` parallel continuous solutions in one forward pass, so one training run yields a *diverse population* of high-quality discrete solutions. QQA4CO ships it under `qqa.pignn.train_cpra_pi_gnn`, sharing the same `GCNNet` backbone and `BinaryRelaxation` penalty as `train_cra_pi_gnn`.\n",
     "\n",
@@ -436,7 +436,7 @@
     "\n",
     "```bibtex\n",
     "@article{ichikawa2025cpra,\n",
-    "  title  = {Continual Parallel Relaxation Annealing for Diverse Solutions in Combinatorial Optimization},\n",
+    "  title  = {Continuous Parallel Relaxation for Finding Diverse Solutions in Combinatorial Optimization Problems},\n",
     "  author = {Ichikawa, Yuma and Iwashita, Hiroaki},\n",
     "  journal= {Transactions on Machine Learning Research},\n",
     "  year   = {2025},\n",

notebooks/cra_pignn_example.ipynb

@@ -470,7 +470,8 @@
     "- **Blackwell / B200 GPU sanity check**: `scripts/sanity_pignn_gpu.sbatch` (Slurm) verifies `sm_100` is in `torch.cuda.get_arch_list()` and runs the demo on the GPU.\n",
     "- **CLI**: `qqa solve --problem mis --backend pignn --learning-rate 1e-3 --pignn-init-reg-param -2 --pignn-annealing-rate 5e-4 --epochs 5000`.\n",
     "- **Reference DGL implementation**: <https://github.com/Yuma-Ichikawa/CRA4CO> (NeurIPS 2024).\n",
-    "- **QQA paper**: Ichikawa & Arai, *\"Continuous Tensor Relaxation for Finding Diverse Solutions in Combinatorial Optimization\"*, ICLR 2025 ([OpenReview](https://openreview.net/forum?id=9EfBeXaXf0))."
+    "- **QQA paper**: Ichikawa & Arai, *\"Optimization by Parallel Quasi-Quantum Annealing with Gradient-Based Sampling\"*, ICLR 2025 ([OpenReview](https://openreview.net/forum?id=9EfBeXaXf0), [arXiv:2409.02135](https://arxiv.org/abs/2409.02135)).\n",
+    "- **CPRA paper (CRA / CPRA backends)**: Ichikawa & Iwashita, *\"Continuous Parallel Relaxation for Finding Diverse Solutions in Combinatorial Optimization Problems\"*, TMLR 2025 ([OpenReview](https://openreview.net/forum?id=ix33zd5zCw))."
    ]
   }
  ],

pyproject.toml

@@ -25,14 +25,27 @@ keywords = [
 ]
 classifiers = [
     "Development Status :: 4 - Beta",
+    "Environment :: Console",
+    "Environment :: GPU :: NVIDIA CUDA",
     "Intended Audience :: Science/Research",
+    "Intended Audience :: Education",
+    "Intended Audience :: Developers",
     "License :: OSI Approved :: BSD License",
+    "Operating System :: OS Independent",
+    "Operating System :: POSIX :: Linux",
+    "Operating System :: MacOS",
+    "Operating System :: Microsoft :: Windows",
     "Programming Language :: Python :: 3",
     "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
     "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3 :: Only",
+    "Topic :: Scientific/Engineering",
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Scientific/Engineering :: Mathematics",
     "Topic :: Scientific/Engineering :: Physics",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+    "Typing :: Typed",
 ]
 dependencies = [
     # The lower bound matches ``requirements.txt`` (used by Streamlit Community

src/qqa/__init__.py

@@ -1,8 +1,9 @@
 """Quasi-Quantum Annealing (QQA) for combinatorial and spin-glass optimization.
 
 Reference:
-    Y. Ichikawa, Y. Arai. "Continuous Tensor Relaxation for Finding Diverse
-    Solutions in Combinatorial Optimization." ICLR 2025.
+    Y. Ichikawa, Y. Arai. "Optimization by Parallel Quasi-Quantum Annealing
+    with Gradient-Based Sampling." ICLR 2025.
+    https://openreview.net/forum?id=9EfBeXaXf0  (arXiv:2409.02135)
 
 Typical usage::