[MNT] update statsforecast version in forecasting extra

[yarnabrina]

PR #5920 intended to update bounds for statsforecast in all dependency specifications. As noted in #6057, it was done only in all_extras and all_extras_pandas2. This PR updates it for the missed single component optional dependency forecasting.

[fkiraly]

why did dependabot not catch this?

[yarnabrina]

I don't know why, but I noticed it never does for dependencies that have python version markers.

[fkiraly]

strange, but interesting observation

[yarnabrina]

CI took more than 18 hours, but it's finished finally. Ready for review+merge.

[fkiraly]

... why did it take 18 hours???

Any relation to the change made by this PR? Or is it more an instance of "new CI is less optimal in the testing everything case"?

[yarnabrina]

All jobs were queued for about 10 hours. I don't know why, but don't think it's caused by new CI or this PR changes.

is it more an instance of "new CI is less optimal in the testing everything case"?

Total runtime wise, sure, but it checks with different (more up to date) versions so that probably should also be kept in mind. As you explained earlier, job creations and setups takes a very significant time.

[fkiraly]

As you explained earlier, job creations and setups takes a very significant time.

We should somehow address this, for PRs like this. Clogging the CI pipeline for half a day is not good.

[yarnabrina]

Agreed, triggering all tests even though mostly they do nothing but install is the culprit, and will happen for all PR's that modify pyproject.toml even slightly (I think, though I'm not 100% sure on that). Other than waiting for #5477 or #5719, not sure what to do, and those will need some good focused time which I currently lack.