Add llms-full.txt, AGENTS.md, context7.json and a machine-readable API reference

[ru551n]

What this PR does

It makes VUnit's documentation and APIs easy for AI coding assistants to use. The docs build generates these files on the website:

• llms-full.txt: all documentation pages as plain text in one file, including the VHDL package headers that the pages pull in.
• api/vunit_lib.json and api/python.json: the VHDL and Python APIs in machine-readable form. The VHDL part covers subprograms with their parameter modes, types and defaults, and aliases, types, constants and doc comments for all 70 packages. It is extracted from syntax trees dumped by vhdl-dump-ast, from the VHDL-LS project. The docs job installs version 0.1.0 with cargo install.

Two files are added to the repository:

• AGENTS.md: the commands and rules a coding agent needs when contributing: tests, lint, news fragments, the license header and generated files.
• context7.json: configuration for Context7 (see below).

Why it is beneficial

AI assistants write run scripts and testbenches from the documentation they can find, and today they miss the VHDL API. The API pages include the package headers with literalinclude, so a tool that reads the .rst sources never sees them. The result is guessed subprogram names and outdated usage.

With these files, an agent can look up exact overloads, parameter modes and defaults instead of guessing. Nothing needs manual upkeep: every docs build regenerates the files, and the build fails if the syntax tree format changes unexpectedly.

What Context7 is

Context7 is a documentation service that AI coding assistants query for up-to-date library documentation. It already indexes this repository as /vunit/vunit. Asked about queue_pkg's new_queue, push and pop, it currently returns only com user guide examples and a 2018 blog post, because the API headers are not in the .rst sources.

This PR improves that in two ways. context7.json leaves out the blog and release notes and gives usage rules, and llms-full.txt can be added as a Context7 source so that the API text is indexed.

After merging, a maintainer needs to:

1. claim /vunit/vunit on context7.com, which adds a key to context7.json;
2. submit https://vunit.github.io/llms-full.txt as a source.

Testing

The unit tests and lint pass, and tox -e py313-docs passes locally. The generated VHDL API was compared against the output of an earlier GHDL-based version of the generator.

This PR was prepared with substantial use of an AI assistant (Claude Code), under my direction and review.

🤖 Generated with Claude Code