refactor(ci): extract alert formatting into composite action and isolate Python tests

[geovannefduarte]

Description

Three workflows duplicated the same inline shell pattern for formatting CI failure alerts (build JSON with jq, pipe to format_alert.py, write to $GITHUB_OUTPUT).
This PR extracts that pattern into a reusable composite action and moves Python script tests into their own lightweight job so they run in seconds without waiting for the Gradle build pipeline.

Key Changes

• .github/actions/collect-failure-info/action.yml — new composite action that accepts a jobs_json array and context inputs, constructs the alert payload, and calls format_alert.py. Follows the same caller-must-checkout pattern as setup-build.
• .github/workflows/ci-manual-trigger.yml — split the monolithic format-alerts step into a job-filtering step and a composite action call, removing ~20 lines of inline shell.
• .github/workflows/periodic-green-check.yml — replaced inline staleness alert formatting with the composite action.
• .github/workflows/post-alerts.yml — replaced inline test alert formatting with the composite action.
• .github/workflows/build-and-test.yml — added a standalone python-tests job (no JVM/Gradle, runs in parallel) and removed Python test steps from the Gradle test matrix job, eliminating 18 redundant step executions per CI run (3 steps x 6 matrix cells).

How was it tested? What documentation was provided?

CI and gradle build

• actionlint passes on all modified and new workflow files
• Python unit tests pass locally via python3 -m unittest discover -s .github/scripts/tests
• Alert output format is unchanged (composite action produces identical JSON payloads to the inline shell it replaces)

[rstata]

Review

Good refactoring overall — the composite action is well-structured, output references are preserved correctly across all callers, and isolating the Python tests into their own job (15s standalone vs duplicated across 6 matrix cells) is a nice efficiency win.

One bug and one nit.

Bug: multi-line jobs_json will break $GITHUB_OUTPUT in ci-manual-trigger.yml

Lines 72/75/78 call jq without -c, so the output is pretty-printed (multi-line). Line 80 then writes it as:

echo "jobs_json=$jobs_json" >> "$GITHUB_OUTPUT"

The echo "key=value" format only works for single-line values. When a CI job actually fails, the multi-line JSON will be truncated at the first newline, and the composite action will receive invalid JSON.

CI is green on this PR because the format-alerts job only runs on failure — this path was not exercised.

Fix: add -c (compact) to the jq calls:

jobs_json=$(echo "$jobs_json" | jq -c --arg name "Build and Test" --arg run_id "$BAT_RUN_ID" '. + [{"name": $name, "run_id": $run_id}]')

(Same for the SDT and BCV lines.)

Nit: display name change

"Standalone Tests" was renamed to "Demo App Tests" on line 75. Intentional? Fine either way, just flagging since it'll change the alert text.

[rstata]

from claude:

Bug found: ci-manual-trigger.yml line 80 writes multi-line jq output (pretty-printed JSON) to $GITHUB_OUTPUT using the single-line echo "key=value" format. This will silently truncate the jobs_json value when a CI job actually fails, feeding invalid JSON to the composite action. The fix is adding -c to the three jq calls so they produce compact single-line JSON. This wasn't caught by CI because the format-alerts job only runs on failure.

If you confirm this is a bug is there a regression test you can add for it?