Since #135113, calls to /im_livechat/get_session can become very expensive on large databases with a long history of livechat sessions. Operators who have thousands of past livechat sessions (channels) can cause the SQL query in get_operator() to take multiple seconds.

When the livechat is set to auto-popup and there are lots of visitors, this can become very significant.

This patch changes two aspects:
- introduce a CTE for the RTC session part, in order to avoid a JOIN cardinality explosion between livechat channels and channel_members - we only care about channel membership for a RTC session AFAICT
- restrict the selection of livechat channels created **within the last 24h**, considering that we only care about messages sent in the last 30 minutes anyway

The second part may change the result ordering in the presence of old channels (>24h) with recent messages - but that is rather unlikely, and an operator will be selected anyway.

This change makes a huge difference on a database with 300k livechat channels and 1 million `discuss_channel_member` records, from several seconds to >50ms. The results are identical or extremely similar in most cases, and the database only needs to look at a few hundred records instead of millions.

This shouldn't change anything regarding SEO, but is worth a try.
It will also impact the link suggestion when creating a link in the
editor, but having pages listed first also have sense there, or at least
it won't be worst.

The idea for the SEO part is that if the sitemap order (if too long)
would have some impact as crawlers might have a limited crawling budget
for your website and it will only crawl the first pages it finds.
Are those "first pages" impacted by the sitemap order? It's almost sure
it's not, as Google definitely knows how to crawl on its own, and is
even probably ignoring the sitemap most of the time.

Also, pages:
- Are probably always important content since you created manually a page to
  write something, while (some) controllers might just be content you
  care less about. Pages are probably always important while we can't
  say that for controllers.
- Should be fewer in number than controllers most of the time
- Have a lastmod set, as opposed to controllers

For all those reasons, this commit reverse the pages vs controllers
order in the sitemap.

This is coming from our prod where some pages are yet not indexed while
they have been published months ago.