Refactor browser_session table

[ncosta-ic]

This PR modifies the browser_session table to reflect new changes to the data structure.

• The primary key no longer depends on the username or user_agent, as the session storage hook takes care of zombie entries before inserting new session data.
• Added the additional descending order to the existing authenticated_at index.
• Fixed index naming convention.
• A new combined index has been added, which speeds up lookups that use both username and user_agent conditions.
• The default expression for authenticated_at got removed, as the timestamp now gets supplied by the PHP daemon itself.

Fixes Icinga/icinga-notifications-web#194
Fixes Icinga/icinga-notifications-web#195

[julianbrost]

• Reverted the usage of citext on the username property, as this idea has not been finally discussed (see: schema: Require & use citext extension #71).

Is that really due to that? Is there any remaining doubt that the username can be treated case-insensitive? To my current knowledge, there seems to be some inconsistency in Icinga Web there between MySQL/MariaDB and PostgreSQL, but neither allows you to have users whose names differ only in case.

[ncosta-ic]

Is that really due to that? Is there any remaining doubt that the username can be treated case-insensitive? To my current knowledge, there seems to be some inconsistency in Icinga Web there between MySQL/MariaDB and PostgreSQL, but neither allows you to have users whose names differ only in case.

That's correct. We don't support different usernames solely based on their case sensitivity.
However, I feel the introduction of citext should take place in a separate PR once everything's discussed about it.
I'd rather not include in the browser_session if that's not our standard yet.

[julianbrost]

#71 was merged earlier today so now there's a precedent for using citext to store the username.

[ncosta-ic]

@julianbrost I rebased this PR and reverted the username changes.
Feel free to merge it, if that's fine for you now :)