JOSS review: contributing/support guidelines and a few install and config commands #358

This is part of my review in openjournals/joss-reviews#11291 (the "Community guidelines" and "Installation instructions" items). Everything below is as of current main (b39657a).

Community guidelines

JOSS asks for clear guidance on three things: how to contribute, how to report problems, and how to get support. At the moment:

• CONTRIBUTING.md is only about linting (ruff, pre-commit). It doesn't describe the process for outside contributors: fork/branch/PR, running tests, review, and any licensing terms for contributions to LANL code.
• I couldn't find anything in the README, CONTRIBUTING.md or the docs that tells users where to report bugs or where to ask for help. The Issues link is only in the PyPI metadata, and Discussions are turned off.
• There are three contributing pages (CONTRIBUTING.md, docs/contributing.md, docs/best-practices/contributing.md) with different content. The last one reads like a generic template ("If the project defines dependency groups...").

A short "Reporting issues" and "Getting help" section plus a few lines on the PR process would cover it. Since the software runs code on the user's machine and has a web UI, a SECURITY.md would also be good to have.

Install commands that don't work as written

• CONTRIBUTING.md and docs/contributing.md: pip install -e .[dev]. There is no dev extra. dev is a dependency group in pyproject.toml, so pip installs the package without the dev tools and pre-commit is missing.
• docs/agents/lammps.md (lines 27–28) and docs/configuration/models.md (lines 212 and 221) use ursa[lammps] / ursa[dashboard]. The package on PyPI is ursa-ai, so these should be ursa-ai[...].
• It would help to have one table of the optional extras (dashboard, lammps, opt, mp, image, ...) and the system dependencies some of them need (for example lmp_mpi and MPI for the LAMMPS agent).

The Ollama example in docs/configuration/models.md (lines 76–91)

I used this example to run URSA with a local model and hit two problems:

• ollama pull gpt-oss-20b fails, because there is no model with that name in the Ollama library. The name is gpt-oss:20b, so the config should say model: ollama:gpt-oss:20b. That form is accepted on v0.17.1 and main. On v0.17.0 it was rejected by the config validator.
• The base_url set under inference_providers is not used for Ollama. With base_url: http://127.0.0.1:11999, ursa --log-level debug exec "hi" still connects to 127.0.0.1:11434 (v0.17.1 and main). So the example only works when Ollama runs on the default local port, and fails for Ollama on another host, for example from inside a Docker container. Setting base_url directly under llm_model works. I see that Fix Config Merge #352 ("Fix Config Merge") touches the base_url handling. If it fixes this, a note here is enough.