Show pinned model, schema and tools in llm templates

[eren-YU]

Closes #1678.

What changed

llm templates showed only the name and prompt text, so the things a template can pin — model, schema_object, tools, functions, fragments — were invisible from the listing. Answering "which of my templates pin gpt-4o?" or "which ones need tools?" meant running llm templates show on each one, and after a plugin upgrade the templates pinning a specific model are exactly the ones that need attention.

The listing now appends a compact marker suffix, only for the fields a template actually sets:

cmd    : system: reply with macos terminal commands only, no extra information
dog    : invent a dog [schema]
frag   : with fragments [fragments: 2]
report : system: Write a QA report [model: gpt-4o] [tools: llm_time, search]

Markers are omitted when the corresponding field is unset, so a prompt-only template's line is byte-identical to before.

This reaches the templates list path through template_list_markers(); the template object was already loaded in templates_list(), so there is no extra loading or I/O.

How I checked

Two tests added to tests/test_templates.py:

• test_templates_list_markers — sets a pinned model, a schema, tools and fragments, and asserts the full listing output, including the fragment/query-free cases from the issue.
• test_templates_list_markers_absent_for_simple_templates — asserts a prompt-only template still prints exactly one : template one, so the markers cannot leak into existing listings.

The first fails without the change; both pass with it. The existing test_templates_list and test_templates_list_no_templates_found are untouched and still pass.

uv run pytest tests/test_templates.py -q

→ 48 passed. ruff check and black --check are clean.

Note on the alternatives in the issue

The issue also floated a --verbose flag and a separate llm templates --json. I went with the always-on markers because they are small enough to read at a glance and keep list useful without a flag; happy to gate them behind --verbose instead if you would rather keep the default output stable.