[Blueprints] Import WXRs via the DataLiberation importer (first stab)

[adamziel]

Adapts the import-markdown-directory.php script from the create-wp-site tool for ImportContentStep to start using the Data Liberation importer in Blueprints v2.

For example, this Blueprint would use the Data Liberation importer:

{
	"version": 2,
	"content": [
		{
			"type": "wxr",
			"source": "https://raw.githubusercontent.com/wordpress/blueprints/trunk/blueprints/stylish-press/site-content.wxr"
		}
	]
}

In this initial exploration:

• The runner puts php-toolkit.phar in the target site directory
• The step handler creates the import-markdown-directory.php script in the target site directory
• The step handler buffers the entire WXR file into the target site directory
• The step handler runs the import-markdown-directory.php script in a subprocess, pointing it to the buffered WXR file

Remainig work work

• Source php-toolkit.phar from somewhere in a production blueprints.phar release. GitHub Releases maybe?

Follow-up work

• Remove Data Liberation files from the blueprints.phar build to trim it down to ~400KB
• Pass configuration options to the importer, e.g. allowed media domains, author mapping mode etc.
• Stream the WXR file directly from its reference. Do not buffer it first. Ditto for the "type": "posts" import mode.
• Use the Data Liberation importer also for importing posts.
• Pass progress updates from the importer script to the Blueprint runner
• Use the same data reference resolution mechanism in Blueprints and in the importer to support sourcing media files from any remote execution context (e.g. a git repo).
  • Idea: Don't process downloads in the subprocess. Instead, use message-exchange to run all the downloads from the parent process.
    • Downside: Message-passing complexity.
    • Upsides: Resource management is centralized. We can resolve things once without worrying about duplication between the parent process and child. No need for exceptions such as "Do not fetch these assets in the runner, let the child process handle that". We'd automatically respect execution context boundaries.