Build augmented prompts for an LLM

Use the prompts module when you need to assemble a grounded prompt from retrieved passages before sending a query to an LLM.

Prerequisites

Format your retrieved hits into a context string. Call join_context() to concatenate hit contents into a single sentinel-wrapped string, or call join_context_numbered() if you want each passage labelled [P1], [P2], and so on inside <passage> tags:
```
from attune_rag.prompts import join_context, join_context_numbered

# Plain sentinel-wrapped context
context = join_context(hits)

# Numbered passage context
context = join_context_numbered(hits)
```
Both functions accept an optional corpus and a max_chars limit to cap total context size.
Build the augmented prompt. Pass your query string and the context string to build_augmented_prompt(). Supply a variant argument to select a prompt style; the default is 'baseline':
```
from attune_rag.prompts import build_augmented_prompt

prompt = build_augmented_prompt(query=query, context=context, variant="baseline")
```
- query must be a non-empty string; otherwise a ValueError is raised.
- If you pass an unrecognised variant, a ValueError lists the valid options.
Send the rendered prompt to your LLM. Pass the string returned by build_augmented_prompt() directly to your model's API. The prompt already includes the injection-defence clause that instructs the model to treat content inside <passage>…</passage> tags as documentation, not as instructions.
Run the prompt-related tests. Verify your integration with:
```
pytest -k "prompts"
```

The task succeeded when:

build_augmented_prompt() returns a non-empty string containing your query and the sentinel-wrapped passages.
All pytest -k "prompts" tests pass with no errors or failures.