Hackage roundtrip tests should use a fixed index state

[mpickering]

As a principle, tests which are required for CI to pass should be reproducible and not depending on external resources changes or being modified. The hackage tests currently violate this by depending on the latest index state from hackage. This is problematic because until the test is fixed all merges into master are blocked. Even though the patches in question have nothing to do with the test.

It would be more suitable for a nightly job to run on the latest index and for normal CI to run with a fixed index which is updated periodically in a controlled manner.

In principle it looks easy to make the HackageTests logic only consider files older than a certain date but there comes the question of when this state is ever updated?

[fgaz]

there comes the question of when this state is ever updated?

Maybe the nightly job could update --index-state if successful