Add test phase in CI for stdexec configuration

[aurianer]

• Add test phase for the stdexec configuration
• Early release of the tiles in unwrap to avoid hangs using std::tuple (previously handled by using pika's version of ensure_started after transform Remove explicit use of futures in tridiagonal solver #658)

[rasolca]

cscs-ci run

[aurianer]

Note: I have a timeout on test_tridiag_solver_distributed, investigating..
EDIT: The hang has been introduced by removing the ensure_started in applyDeflation and calcTolerance. I will look at removing those in a separate PR (issue #954 opened).

[aurianer]

cscs-ci run

[msimberg]

Thanks @aurianer for the investigation here.

Functionally I'm happy with this change but see inline comment before I'm ready to approve this.

Linking #665 as well since it's a related issue. I think the solution here is necessary, but possibly not sufficient for full control of tile lifetimes (as the remaining ensure_started problem hints at).

[aurianer]

cscs-ci run

[msimberg]

This looks good to me now, thanks!

I'd still like to wait to see the benchmark results, but assuming there are no surprises there this requires no further changes.

The missing includes from unwrap.h are not a blocker for this PR.

[aurianer]

So I:

• added the missing include
• change the early release implementation with @msimberg suggestion which fixes the performance regression
• just launched the benchmark with the new version

TODO: Post the benchmark results when available

[msimberg]

cscs-ci run