Document mutation-testing setup for Ralph projects

[rxdt]

Ralph should make agents write tests that challenge behavior, not just chase 100% line coverage.

Create a concise docs/example page showing how to add mutation testing to a Ralph-managed Python project.

Suggested scope:

• use mutmut as the first Python example
• show modern config names such as source_paths instead of deprecated settings
• show the import-path gotcha for projects that use src/ or package modules
• explain how to interpret surviving mutants
• keep it practical: commands, expected output shape, common failure modes

Acceptance criteria:

• a new user can copy the config and run the command
• docs explain why mutation testing catches weak agent-written tests
• no broad framework comparison; keep the page short and actionable

[KUMAR123gh]

Hi! I'd like to work on this issue.

While looking through the repository, I noticed that docs/Phase 4 Launch mentions that the README shouldn't include a demo transcript until:

• uvx ralph-harness demo is available and completes successfully
• the PyPI package is published

Since the acceptance criteria also mention using a real run, I wanted to check whether this issue is waiting on those milestones, or if there's already an expected demo transcript I should use for the README.

Thanks!

[rxdt]

Thanks. This issue is specifically for mutation-testing docs, not the README demo transcript. PyPI is now live, but you do not need to wait on README/demo work for this one.

Good scope for this issue:

• add a concise mutmut setup note for Ralph-managed Python projects
• use current config names, especially source_paths
• include the common import/path gotcha for package or src/ layouts
• show how to run it and how to interpret surviving mutants
• keep it practical and short

The README demo transcript is tracked separately in #3.

[rxdt]

Also: if you want to go beyond docs, you are welcome to open a PR that adds mutmut support directly to ralph-harness as an optional framework feature.

Keep it opt-in and small. The core goal is still the same: make it harder for agent-written tests to pass by only chasing line coverage.