Fix space leak on cradle reloads

[pepeiborra]

We were leaking 'PackageExports' values for all past 'HscEnv's.
Fixed by moving the package exports inside the 'HscEnvEq'.

I have also added two experiments to characterise the leak, the main one is 'code actions after cradle edit'. On the Cabal 3 example, this is what the experiment looks like:

[Ailrun]

LGTM

[wz1000]

Now that the hiedb changes have landed, I think this feature would be a good candidate to implement using the database.

[pepeiborra]

Now that the hiedb changes have landed, I think this feature would be a good candidate to implement using the database.

But hiedb doesn't index build depends, does it?

[wz1000]

But hiedb doesn't index build depends, does it?

Ah, right. I misread the PR and thought it was for the project local exports map.

However, we could still index and persist a export map for dependent packages, by indexing the exports directly from the ModIface instead of the .hie file.

[pepeiborra]

All the tests pass now, but the benchmark run on master (where the leak is still present) times out. I'll wait for the speed ups in #1315 to land

[pepeiborra]

But hiedb doesn't index build depends, does it?

Ah, right. I misread the PR and thought it was for the project local exports map.

However, we could still index and persist a export map for dependent packages, by indexing the exports directly from the ModIface instead of the .hie file.

That's how the exports map works now, except that it isn't persisted. Do you think that persisting would be worth the effort? I think that's a nice idea, please file an issue!