[felipebergamin]

Closes kernelci#2114
Closes kernelci#1080

[REGRESSION] mail used to Cc the default_recipients of whichever subscription file had the issue's git URL, whatever the branch or architecture. The [STATUS] summary Cc's the same list, so a maintainer of one stable branch got regressions from every branch.

This PR moves the regression Cc to backend/data/notifications/regression/. An entry there can narrow who gets Cc'd by branch, architecture, config or compiler. default_recipients in subscriptions/ is still the [STATUS] Cc and is not modified.

Commits

1. copy tree recipients into regression/ is data only. Each of the 45 non-hardware subscription trees gets regression/<tree>.yaml with one unscoped entry, copied from its default_recipients (display names kept). regression/default.yaml holds the To address. Nothing reads these files in this commit, so it can be reviewed by checking that each list equals the old one.
2. Cc regression reports from regression/ contains the logic.

How it works

# regression/stable.yaml
stable:
  - recipients: [stable@vger.kernel.org]   # no scope keys: every build issue on this tree's URL
  - branch: [linux-6.12.y]
    architecture: arm64
    recipients: [maintainer@example.com]

• The top-level key must be a tree key from subscriptions/. The loader uses that tree's url to match issues.
• An entry matches when every key it sets matches. branch is compared with the issue's branch, the one printed in the mail. architecture, config_name and compiler are compared with the build incidents generate_issue_report already fetches, so there is no new query.
• The Cc is the union of matching entries. An address listed twice, even with different display names, is sent once.
• test_report uses only entries without scope keys, which reproduces today's list.
• With --add-mailing-lists, regression reports take To from regression/default.yaml. [STATUS] keeps KERNELCI_RESULTS. Reply-To and the regzbot Cc did not change.
• Build and boot report bodies start with Cc from: regression/<tree>.yaml so a reader knows which file to edit.

typeModels/notificationRecipients.py validates the files. An unknown tree, a repeated subscription URL, platform, origin, any other unknown key, issues other than build, or a *_hardware.yaml file raises an error that names the file, entry index and key. A unit test loads the committed directory, so a bad edit fails pytest -m unit in CI.

send_email_report no longer looks recipients up itself. Callers pass recipients=, so I removed the git_url and signup_folder arguments that only fed the old lookup.

Trade-offs

• The header line uses the issue's tree_name, while the Cc is chosen by URL. If a KCIDB tree name differs from the subscription key, the header names a file that does not exist. Today the keys follow the dashboard tree names.
• test_report also counts as a regression report, so with --add-mailing-lists it takes To from default.yaml as well. The address is the same as before.

Test plan

From backend/:

• poetry run pytest -m unit kernelCI_app/tests/unitTests/commands/ passes, including the new regression_recipients_test.py and the refreshed golden reports.
• Break a file on purpose, for example add platform: foo to an entry in regression/stable.yaml. The real-directory test fails with regression/stable.yaml: entry 0: key 'platform'.
• Dry run without sending: poetry run python manage.py notifications --action issue_report --id <maestro build issue id>. The body starts with Cc from: regression/<tree>.yaml.
• Add --send and answer n at the prompt (do not pass --yes). The printed Cc is the tree's old default_recipients. Add --add-mailing-lists and check that To is kernelci-results@groups.io.
• Add a scoped entry, for example branch: <another branch> with a test address, and repeat. That address is not Cc'd. Change the branch to the issue's branch and it is.
• poetry run python manage.py notifications --action summary --summary-origins maestro without --send behaves as before. [STATUS] still reads default_recipients.

Locally: 970 backend unit tests pass, ruff check and ruff format --check are clean.