Is your feature request related to a problem? Please describe.
I'm trying to get a mental understanding of the computational cost of Alerting rules. To understand if it may be dangerous to add alerts aggregating long time intervals.

I don't understand if the evaluation runs continuously as logs are ingested or if it runs periodically, by re-querying all the logs again.
If it runs periodically, there seem to be two ways to configure it, either in the rules.yaml, each group has a interval, or the ruler section of the configuration has an evaluation_interval parameter.

Describe the solution you'd like
A: Documentation should clearly describe how and when alerting rules are evaluated.
B: Documentation should clearly describe what evaluation_interval does, what groups.interval does and the interaction between them.
C: Documentation for rules should not distinguish Alerting vs Recording rules where they have common behaviour. The interval parameter is only (briefly) mentioned for Recording rules but not for alerting rules, where the example is described as "A complete example of a rules file".

Describe alternatives you've considered
I have carefully read all the following pages and can not connect the pieces :(
https://grafana.com/docs/loki/latest/get-started/architecture/
https://grafana.com/docs/loki/latest/get-started/components/#ruler
https://grafana.com/docs/loki/latest/alert/
https://grafana.com/docs/loki/latest/operations/recording-rules/
https://grafana.com/docs/loki/latest/configure/#ruler

Additional context
Some of the confusion arises from some pages saying that the ruler "continually" evaluate a set of rules. As opposed to "periodically".