Default GH runners run out of disk space when building images for quince environments

[mariajgrimaldi]

Description

This issue of running out of resources when building images with themes configured was initially addressed in PR #12. However, the fix only works for Redwood environments, which build assets differently than in previous versions: https://openedx.atlassian.net/wiki/spaces/COMM/pages/3890380898/Current+Release+Redwood+-+Operator+Dev+Notes

I believe this change in how assets are managed affects the building process making it more expensive resource-wise. Here's the error we're getting when building quince images: https://github.com/eduNEXT/ednx-strains/actions/runs/11411669740, while redwood works fine https://github.com/eduNEXT/ednx-strains/actions/runs/11410021825/job/31755748137 (see step Build service image with no cache)

I haven't tested it with previous versions with configured themes, but it's probably also affecting them.

How to reproduce

1. Run a new build action in ednx-strains.
2. For the workflow configuration set MJG/releases-gh in the "Use workflow from" field.
3. Use MJG/release-gh to test the errors.
4. The build should finish successfully with redwood (redwood/base as ednx-strain path)
5. The build should stop or get stuck when building with quince (quince/dedalo). If it doesn't stop, please cancel before reaching the 40min mark.

Possible solutions

1. Use a fixed runner for all builds
2. Use on-demand self-hosted runners: continue implementation of configurable self-hosted runners previously discussed in feat: use self-hosted runners to improve build performance #6 feat: self hosted runners with ec2-action-builder #7
3. Use more powerful GH runners