fix: limit depth of index discovery to prevent slow scans

[majiayu000]

Summary

• Fixes the issue where leann list scans all files in large directories like $HOME
• Adds _find_meta_files_limited() method with configurable max_depth (default: 3)
• Skips common large directories (node_modules, .venv, .git, pycache, etc.)
• Applies limited search in both CLI and registry modules

Fixes #122

Test plan

• Added 8 new tests in tests/test_cli_list_performance.py
• Tests verify depth limiting works correctly
• Tests verify skip directories are properly excluded
• All new tests pass locally

[andylizf]

Thanks for working on this! A few suggestions:

1. Add --max-depth CLI option

Instead of hardcoding max_depth=3, consider adding a CLI option with a reasonable default:

leann list --max-depth 5

3 levels might be too shallow for some project structures (e.g., project/data/experiments/v1/index.leann would be missed).

2. Consider a design improvement: Full registry instead of scanning

Currently the code scans for "Apps format" indexes (*.leann.meta.json) because they can be anywhere in the project. But this is the root cause of the performance issue.

A better approach might be to register all indexes (both CLI and App-created) in the global registry with their full paths:

// ~/.leann/indexes.json
{
  "my-docs": "/path/to/project/.leann/indexes/my-docs",
  "app-index": "/path/to/project/data/app-index.leann"
}

This way leann list becomes O(1) - just read the registry, no scanning needed at all.

This could be a follow-up issue/PR, but worth considering as the long-term solution.

[majiayu000]

Thanks for the feedback @andylizf!

I've added the --max-depth CLI option as suggested. Users can now customize the scan depth:

# Default depth (3)
leann list

# Scan deeper directories
leann list --max-depth 5

For the global registry suggestion (making leann list O(1)), I agree it's a great long-term improvement. I'll create a follow-up issue to track that enhancement.

[majiayu000]

Update: I've also implemented the long-term global registry solution in a separate PR #199.

• This PR (fix: limit depth of index discovery to prevent slow scans #196) - Short-term fix: adds --max-depth CLI option
• PR feat: implement global index registry for O(1) index discovery #199 - Long-term solution: implements O(1) registry lookup (closes feat: implement global index registry for O(1) index discovery #198)

This PR can be merged independently. PR #199 builds on top of this one.