Separate definition of task dependencies

Right now for very large `.cirrus.yml` configurations there is a problem of dependency definition. All the `depends_on` are sparsed across all the task definition and it might be hard to wrap your head around it. What if we'll introduce a new way to define the execution graph via a top-level `dependecies` definition?

Here is a configuration example of a continious deployment:

```yaml
# .cirrus.yml

dependencies:
Deploy:
only_if: $CIRRUS_BRANCH == "main"
requires:
- Check Incidents
- Push

task:
name: Push

task:
name: Check Incidents
script: ... # checks that there is no active incident

task:
name: Deploy
```

In case `dependencies` is used, parser will error if `depends_on` or `only_if` are used on tasks so there is strictly two different options to specify dependencies without combining them. This way with the new `dependencies` we can change behaviour slightly. For example, treat `requires` as a strict `depends_on`.

Creating this ticket as a brain dump and a conversation starter.

No source files or tests are named. Start by locating the .cirrus.yml parser and existing handling for task-level depends_on and only_if; define the top-level dependencies syntax, validation rules, and strict requires behavior, then add coverage for the new configuration and parser errors.

Written by the indexing model from the issue text.