End to end and integration tests should be run for a controlled matrix of compiler releases

The end to end and integration tests use the Swift compiler provided by the GitHub CI runner. The runner image was updated to include Swift 6.1 (https://github.com/actions/runner-images/blob/ubuntu24/20250406.1/images/ubuntu/Ubuntu2404-Readme.md) but this caused the end to end and integration tests to fail because they install the 6.0.2 release of the Static Linux SDK, which is not compatible with Swift 6.1.

#79 fixes this problem in the short term by updating the workflows to a compatible version of the Static Linux SDK, but ideally these tests should not depend on a the particular version of Swift in a runner image, and should also test multiple versions.

These tests do not run in containers because part of the test involves building and running a container image, which is inconvenient when the test itself runs in a container, as used for the matrix unit tests.

Possible solutions:
* run the end to end and integration tests in privileged containers and pass the Docker socket through so images can be built and run
* run the tests in Docker-in-Docker, which can make it easier to set up and tear down such recursive containers.
* Use named runner images with known Swift compilers instead of `ubuntu-latest`.
* Install specific compilers using `swiftly`.

Start by reading the GitHub CI workflows for the end-to-end and integration tests and the changes in #79. Compare the listed approaches—privileged containers, Docker-in-Docker, named runner images, or swiftly-installed compilers—and determine how a controlled compiler matrix can still build and run container images. Done means these tests no longer depend on ubuntu-latest's Swift version and cover multiple compatible compiler releases.

Written by the indexing model from the issue text.