feat: add configurable local data directory with stale lock recovery

[rodrwan]

Description

Add support for a configurable local data directory, replacing the hardcoded ~/.cie/data/ path. Users can now control where CIE stores indexed data via the indexing.local_data_dir config field or the CIE_DATA_DIR environment variable. Relative paths are resolved from the directory containing .cie/project.yaml.

Additionally includes stale RocksDB LOCK file recovery (auto-removes orphaned locks from crashed processes), graceful shutdown improvements, and build system fixes.

Related Issue

N/A

Type of Change

• New feature (non-breaking change that adds functionality)
• Documentation update
• CI/CD or build changes

Testing

• I have tested this locally
• I have added/updated tests that prove my fix/feature works
• New and existing tests pass locally with my changes

Tests added

• cmd/cie/paths_test.go — unit tests for projectDataDir, dataRootFromConfig, path resolution (absolute, relative, default fallback)
• pkg/storage/embedded_test.go — TestIsStaleLock_NoLockFile, TestIsStaleLock_StaleLock, TestIsStaleLock_ActiveLock

Checklist

• My code follows the project's code style
• I have run make fmt and make lint
• I have added comments for complex code sections
• I have updated documentation if needed
• My changes generate no new warnings

Key Changes

Configurable data directory

Method	Example	Behavior
Config field	indexing.local_data_dir: "/mnt/cie-data"	Absolute path used as-is, /<project_id> appended
Config field (relative)	indexing.local_data_dir: ".cie-data"	Resolved from project config directory
Env var	CIE_DATA_DIR=/mnt/cie-data	Overrides config field
Default	(unset)	~/.cie/data/<project_id>/ (unchanged behavior)

Legacy data migration warning

When local_data_dir is configured and data exists at the old default path (~/.cie/data/<project_id>/) but not at the new path, CIE prints a warning guiding the user to re-index.

Stale RocksDB LOCK recovery (pkg/storage/embedded.go)

If CozoDB fails to open and a LOCK file exists, CIE checks whether any process actually holds the flock. If not (crashed/killed process), it removes the stale lock and retries. This eliminates the need for users to manually delete lock files after crashes.

Graceful shutdown improvements

• All signal handlers now call signal.Stop(sigChan) so a second Ctrl+C force-quits immediately
• User-facing message: "Shutting down... Press Ctrl+C again to force quit."
• Applied to: index, mcp, and serve commands

Build fixes (Makefile, pkg/cozodb/cozodb.go)

• INSTALL_DIR uses $(GOPATH)/bin instead of $(HOME)/go/bin
• macOS: ad-hoc codesign after install (codesign --force --sign -)
• CGO LDFLAGS: -L${SRCDIR}/../../lib so go install ./cmd/cie finds the vendored static library

Refactoring

• Centralized path resolution in new cmd/cie/paths.go (replaces scattered filepath.Join(homeDir, ".cie", "data", ...) calls)
• Moved checkLocalData logic into LocalPipeline.FunctionCount() — queries through the already-open DB instead of opening a second connection
• storage.EmbeddedConfig.DataDir is now passed explicitly instead of being computed internally

Additional Notes

• Backward compatible: when local_data_dir is not set, behavior is identical to before
• The EmbeddedConfig.DataDir field replaces the internal ~/.cie/data/<project> computation in storage.NewEmbeddedBackend, giving callers full control over the path

[frperezr]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance. Tests pass across all packages.