Refactor service agent management

[juliocc]

This PR significantly improves how Fabric handles service agents. This is a very large PR but IMO it will simplify things moving forward.

The main drivers for this PR are

1. There many more service agents today than when we first designed service agents support in Fabric
2. There are now APIs with more than one service agent
3. New service agents are being added constantly

This PR introduces the following changes:

1. Automated Service Agent List: A new script generates the list of service agents by parsing the official GCP documentation. This replaces the manual maintenance of service-agents.yaml. The idea is to run the script manually from time to time
2. Separate Service Account Outputs: The project.service_accounts output is split into project.service_agents and project.default_service_accounts. The Cloud Service Service Agent is handled like any other service agent within service-agents.yaml.
3. Standardized Terminology: The codebase now consistently uses the term "service agent", removing outdated terms like "robots" and "service identity"..
4. Simplified CMEK Dependency Handling: Legacy CMEK dependency handling within the project module is removed. This logic is now the responsibility of users who require it.
5. New Variable new top-level variable (service_agents_config) to control the default behaviour in regards to service agent handling. By default we enable:
   • Creation of primary service agents
   • Granting default permissions to service agents for any enabled API.
6. Renamed Variable: The project.service_encryption_key_ids variable is renamed to project.service_encryption_key_ids for clarity. (Suggestions for a better name are welcome!)

There's a lot of noise in this PR as a lot of modules, blueprints and tests changed. For reviewers, please focus on the changes within the modules/project directory. While many files have been modified, the core logic changes reside there.

[ludoo]

This is really great. Dropped a couple of comments.

One more poitn we should maybe tackle is inter-service dependency: one example I'm currently facing in real life is Dataproc requiring KMS grants on the GCE, GCS, and its own service agent.

[juliocc]

Waiting for e2e to finish before merging