Investigate options to revamp testing workspace setups

[blorente]

Currently, the mechanism we use to test WORKSPACE customizations is the tesd ./test_version.sh, which populates a template workspace and executes Bazel tests there.

Recently, we have needed to add some customization to the twitter_scrooge implementation (#1050). While designing integration tests for that change, we realized that it's hard to do in an extensible way: #1050 (comment)

We should investigate how to implement more robust integration testing. It could be that we need to use bazel-integration-testing, or a different thing.

Relatedly, as we migrate more things to toolchains, it might make things easier to test. (ref: #940)