CSS cache invalidation fails for remote/production sites after push

[tomatitito]

Problem

After elementor-cli push to a production site, the page still serves stale CSS/content despite the CLI reporting "✔ Invalidated CSS cache". Manual intervention via SSH is required to actually flush the cache.

Root Cause

Two issues in the current invalidation logic:

1. invalidateCss() sets meta to empty string instead of deleting it

In src/services/wordpress-client.ts (L164-176), the method sets _elementor_css and _elementor_element_cache to "" (empty string) via REST API PUT. Elementor does not treat an empty string the same as a deleted/absent meta field — it needs the meta to be fully removed to trigger CSS regeneration.

What works: wp post meta delete <page-id> _elementor_css
What does not work: Setting _elementor_css to "" via REST API

The fix should use null instead of "" to delete the meta via REST API:

meta: {
  _elementor_css: null,
  _elementor_element_cache: null,
}

2. No wp elementor flush-css / wp cache flush for remote (SSH) sites

In src/commands/push.ts (L360-369), the container-based wp elementor flush-css only runs for sites with a container config (local Docker/Podman). Remote/production sites only get the (broken) REST API invalidation.

The CLI already supports SSH for db dump (--ssh and --ssh-path flags). The same SSH config should be usable for running wp elementor flush-css and wp cache flush on remote sites after push.

Steps to Reproduce

1. Configure a production site without container config in .elementor-cli.yaml
2. elementor-cli pull <page-id> --site production
3. Make changes to the local page data
4. elementor-cli push <page-id> --site production
5. Observe: CLI reports success + "✔ Invalidated CSS cache"
6. Load the page in browser (even with hard refresh) — old content/CSS is still served

Workaround

Manually run via SSH after pushing:

ssh user@host "wp --path=/path/to/wp post meta delete <page-id> _elementor_css"
ssh user@host "wp --path=/path/to/wp elementor flush-css"
ssh user@host "wp --path=/path/to/wp cache flush"

Proposed Solution

1. Change invalidateCss() to delete meta fields (set to null) instead of empty string
2. Add optional SSH config to site definitions in .elementor-cli.yaml:

   sites:
     production:
       url: https://example.com
       username: admin
       appPassword: "..."
       ssh:
         connection: user@host
         wpPath: /path/to/wordpress

3. After REST API invalidation, if ssh config is present, also run wp elementor flush-css and wp cache flush via SSH
4. The regenerate-css command should also support the SSH flush path