security: keep indexed files inside the cloned repository

[harshitethic]

Problem

RepoPilot recursively indexes files from cloned repositories and later reads those files into search/AI context.

The existing iter_files() implementation trusted Path.is_file(). That follows file symlinks, so a repository containing a symlink to a file outside the clone could cause RepoPilot to index and read host-side content.

Example:

repo/
  leaked.py -> /home/user/private.py

Because RepoPilot sends selected repository context to the local model, this violates the intended repository boundary even though cloned project code is never executed.

Fix

• resolve each candidate path before indexing;
• require the resolved target to remain under the resolved repository root;
• reject paths that fail resolution or escape the clone;
• use the validated resolved file for size checks and downstream reads;
• keep the existing skip-directory, file-size, and extension filters unchanged.

Tests

Adds regression tests proving:

• a symlink whose target is outside the repository is excluded;
• a normal source file inside the repository is still indexed.

The backend CI job now runs the unittest suite in addition to bytecode compilation.

Scope

This protects RepoPilot's read/index boundary. It does not add code execution or change the local-only Ollama behavior.