Restructure JSON output to organize by system and fix byName collisions

[srid]

Currently, the JSON output produced by devour-flake has two issues:

1. Name collisions in byName index: When building for multiple systems (via --override-input systems), packages with the same name from different systems overwrite each other in the byName mapping. Only the last system's package survives.

2. Flat structure lacks system organization: The current output doesn't clearly separate which outputs belong to which system, making it difficult to consume the data programmatically.

Current JSON structure:

{
  "outPaths": [...],
  "byName": {
    "hello": "/nix/store/...-hello-aarch64"  // x86_64 version was lost!
  }
}

Proposed JSON structure:

{
  "x86_64-linux": {
    "outPaths": [...],
    "byName": {
      "hello": "/nix/store/...-hello-x86_64",
      "firefox": "/nix/store/...-firefox-123"
    }
  },
  "aarch64-linux": {
    "outPaths": [...],
    "byName": {
      "hello": "/nix/store/...-hello-aarch64",
      "firefox": "/nix/store/...-firefox-123"
    }
  }
}

Benefits:

• No more name collisions - each system has its own byName namespace
• Clear system organization makes it easier to process outputs programmatically
• Sorted outPaths within each system for predictable ordering
• The text output (default) remains unchanged, continuing to list all paths across all systems

Additional improvements:

• Sort outPaths lexicographically for consistent output