feat(cli): add per-project database configuration via centralized config

[j0k3r-dev-rgl]

📋 Pre-flight Checks

• I have searched existing issues and this is not a duplicate
• I understand this issue needs status:approved before a PR can be opened

🔍 Problem Description

Engram uses a single global database (~/.engram/engram.db) for all projects and conversations. When working across multiple unrelated projects, all memories get mixed together in the same DB — making context less relevant and harder to scope.

There's no way to configure project-specific memory isolation without manually setting ENGRAM_DATA_DIR before every session, which is cumbersome and error-prone.

💡 Proposed Solution

Add a centralized project config at ~/.engram/config.json that maps working directories to dedicated databases.

When engram mcp (or any command) starts, it reads the CWD and looks it up in the config. If there's a match, it uses that project's database instead of the global one. If no match is found, it falls back to the global ~/.engram/engram.db.

New config format (~/.engram/config.json)

{
  "projects": [
    {
      "path": "/home/user/my-app",
      "data_dir": "/home/user/my-app/.engram"
    },
    {
      "path": "/home/user/work/client-x",
      "data_dir": "/home/user/work/client-x/.engram"
    }
  ]
}

New CLI commands

# Add current directory to project config (sets data_dir to CWD/.engram)
engram project add

# List all configured projects
engram project list

# Remove current directory from project config
engram project remove

Resolution priority (highest → lowest)

Priority	Source	Notes
1	ENGRAM_DATA_DIR env var	Existing — unchanged
2	~/.engram/config.json project match	New
3	~/.engram global fallback	Existing — unchanged

Behavior details

• Match is exact — CWD must equal project.path
• If data_dir doesn't exist → created automatically on first use (.engram/ inside the project)
• DB filename stays engram.db — only the directory changes
• engram project add reads CWD and sets data_dir to CWD/.engram
• engram project add is idempotent — running it twice does not duplicate the entry

📦 Affected Area

CLI (commands, flags)

🔄 Alternatives Considered

Setting ENGRAM_DATA_DIR per project via shell aliases or .envrc files works, but requires manual setup per machine and isn't portable or discoverable across teammates.

📎 Additional Context

This follows the same pattern as other tools that support per-project config (e.g. .nvmrc, .tool-versions, opencode.json) — a single place to configure behavior without per-machine setup.

The implementation touches only cmd/engram/ — no changes to internal/store/ are required since store.New already accepts an arbitrary DataDir.

Files expected to change

File	Change
cmd/engram/project.go	New file — EngramConfig, ProjectEntry structs, cmdProject* functions, config read/write helpers
cmd/engram/project_test.go	New file — unit tests for all new behavior
cmd/engram/main.go	Add osGetwd seam, project config resolution in startup flow, case "project" dispatch, updated usage text