docs: HPC OS compatibility audit for chemsmart agent pipeline

[Hongjiseung-ROK]

Summary

• add a research-only HPC OS compatibility audit at docs/research/hpc_os_compatibility.md
• audit the chemsmart agent pipeline at commit 20cbcdb9 against common HPC login-node OS families
• document OS lifecycle, GLIBC, login-node policy, egress, TUI, filesystem, locale, and environment-strategy findings

Scope

• research-only change
• no source code or test modifications
• target pipeline audited: 20cbcdb9

Notes

• bin/plan.md was intentionally skipped per task instructions because it does not exist
• report concludes the stack is broadly ABI-compatible on modern Linux, but operationally amber on shared login nodes

Validation

• not run (docs-only research task)

[gemini-code-assist]

Code Review

This pull request adds a detailed HPC OS compatibility audit for the chemsmart agent, covering dependency layers, GLIBC compatibility, and login-node constraints across various Linux distributions. Feedback indicates that the binary compatibility section should include additional compiled dependencies like rdkit and pymatgen. Furthermore, a discrepancy was identified between the dependencies listed for the audited commit and those currently present in the repository.

[gemini-code-assist]

The binary compatibility analysis in Section 1.7 appears to omit several major compiled dependencies present in the project's pyproject.toml, such as rdkit, pymatgen, pandas, scikit-learn, pillow, and spyrmsd. Since these packages contain compiled extensions (C/C++/Rust), their ABI compatibility is critical for a comprehensive HPC OS audit. rdkit, in particular, is known for having complex dependency requirements that can be challenging in restricted HPC environments.

Suggested change

	Packages that matter for binary compatibility:
	- `pydantic-core`
	- `numpy`
	- `scipy`
	- `PyYAML`
	- `watchdog`
	- possibly `jiter` pulled by current `openai` / `anthropic`
	Packages that matter for binary compatibility:
	- `pydantic-core`
	- `numpy`
	- `scipy`
	- `rdkit`
	- `pymatgen`
	- `pandas`
	- `scikit-learn`
	- `PyYAML`
	- `watchdog`
	- `pillow`
	- `spyrmsd`
	- possibly `jiter` pulled by current `openai` / `anthropic`

[gemini-code-assist]

There is a significant discrepancy between the dependencies described in this audit (for commit 20cbcdb9) and the current state of the repository (36f13c6e). Specifically, pydantic, openai, anthropic, python-dotenv, textual, watchdog, and pyperclip are missing from the pyproject.toml and environment.yml files provided in the current worktree. It would be helpful to clarify if these dependencies were removed or if they are expected to be added in a separate PR, as this may confuse users trying to deploy the current head based on this audit.