[Suggestion]: index files inside linked git worktrees that live under repo root (like .worktrees/)

### Which fff frontend(s)?

MCP server (fff-mcp)

### What problem are you trying to solve?

## In short

I use `fff-mcp` daily and I am very thankful for it — it became part of my hands. But there is one case where it cannot help me, and I think it is worth fixing.

When I start `fff-mcp` at the root of repository, it does not see files that live in **linked git worktrees nested under this root**. Typical layout in our team is `/.worktrees//`. We keep `.worktrees/` in `.gitignore` on purpose, so that `git status` stays clean. And here is the whole problem: fff respects `.gitignore` and `.git/info/exclude` honestly, so these files become invisible.

## My environment

- Tool: `fff-mcp` (the MCP build of fff), one positional argument `[PATH]`
- Host: Claude Code. It starts MCP server once per session, and sets working directory of server equal to directory from where editor was launched.
- OS: macOS (darwin)

## What happens now

1. `fff-mcp` decides its scan root through `Repository::discover(&base_path)` → `repo.workdir()` (you can see it in `crates/fff-mcp/src/main.rs`). When I am already inside a worktree, this resolves to that worktree workdir, and then everything works perfectly — so **per-worktree start is not a problem**.
2. But when I start from the main checkout, the root becomes the main workdir. Nested worktrees under `.worktrees/` are thrown away, because they are written in `.gitignore` (and frequently in `.git/info/exclude` too), and fff obeys both of them.
3. In CLI there is no `--no-ignore`, and in `FffCreateOptions` (`crates/fff-c/include/fff.h`) there is no field to switch ignore off. `PATH` is single — so no second root either.

So the conclusion is simple and a little sad: a session that started at the main checkout will **never** see the worktree files, and one instance cannot serve the main checkout and the worktrees together.

## Why this is important for me (and possibly not only for me)

Worktrees nested under the repo (`/.worktrees//`) are not exotic —
for teams that hold many branches checked out at the same time. We put `.worktrees/` into `.gitignore` exactly to keep `git status` quiet, and exactly this gesture is what hides them from fff. Very often I open one session at the root and then walk into a worktree in the middle of work — and from this moment fff, nailed to the main root since launch, becomes blind to the very files I am editing. It is a pity, because the engine itself is fast and good.

### Proposed solution

_No response_

Start in crates/fff-mcp/src/main.rs at Repository::discover(&base_path) and follow how repo.workdir() establishes the scan root and how ignored paths are handled. Review crates/fff-c/include/fff.h for the available FffCreateOptions, then verify that a session started at the main checkout can index files in nested linked worktrees while preserving normal ignore behavior.

Written by the indexing model from the issue text.