feat: support .codegraphignore for excluding files from index

[Krislu1221]

Motivation

CodeGraph currently indexes every file in the project (aside from what .gitignore excludes). In many real-world projects this means significant noise:

• Monorepos with .sandbox-home/, dist/, build/ directories
• Python projects with .venv/, venv/, __pycache__/
• Node projects with node_modules/
• Large test fixtures or generated assets that should not be in the code graph

Users do not want to add these to .gitignore because they are either git-tracked or serve a different purpose. They want CodeGraph to skip them independently.

In my case, 97% of the 7175 indexed files were third-party dependencies under .sandbox-home/ — making codegraph context and codegraph query results dominated by noise.

Solution

Add .codegraphignore support using the same gitignore syntax (the ignore npm package is already a dependency). Patterns from .gitignore and .codegraphignore are OR-d together — a file is excluded if it matches a pattern in either file.

Changes

1. scanDirectoryWalk (filesystem fallback path): loadIgnore() now reads both .gitignore and .codegraphignore from each directory level. When both exist, patterns from both are merged into one matcher.

2. scanDirectory (git fast path): Applies root .codegraphignore as a second-pass filter after git ls-files. Since the git path already respects .gitignore via git ls-files, only .codegraphignore patterns need to be checked additionally.

3. scanDirectoryAsync: Same treatment as scanDirectory.

Usage

Create a .codegraphignore file in the project root (or any subdirectory for the walk path):

# .codegraphignore — CodeGraph-specific ignores
# Uses standard .gitignore syntax

# Python virtual environments
.venv/
venv/
*.venv/*

# Node.js dependencies
node_modules/

# Build artifacts
dist/
build/
__pycache__/

# Sandbox directories
.sandbox-home/

Then run codegraph index --force to rebuild the index with the new exclusion rules.

Testing

• Builds successfully (tsc + copy-assets)
• Existing test suite: 31/35 test files pass (710/778 tests). The 6 failures are pre-existing flaky tests unrelated to this change (MCP roots timeout, framework integration tests).
• Manual verification: created .codegraphignore with .sandbox-home/ pattern in a test project, confirmed the directory is excluded from codegraph files --json output.