Switch the templates to use the invocation API

[jmsmkn]

The invoke container API was added in #4469. It would be good if this is now made the default for templates so that algorithms and methods created this year are future proof.

[chrisvanrun]

Is the idea that the templates switch behaviour based on label or env? As far as I understood, the idea was that an algorithm image is either 'exec' or 'invoke' but not both. And since 'invoke' is slightly more complex we'd add a switch somewhere on the phase level to create algorithms to go either one. So a challenge (phase) would be all 'invoke' or all 'exec'.

Not sure how 'simple' we can keep an algorithm that supports both.

[jmsmkn]

No need to support both, the template only needs to provide the invocation API. That is the most useful for running algorithms as it avoids reloading the model on subsqueuent invocations, and works anywhere in the system. The exec method will not be deprecated, it's just not as useful to have algorithms that implement it.

The API method is identified with an image label, that is it. It defaults to exec. In a few years we could change that default to invoke.

[jmsmkn]

And since 'invoke' is slightly more complex we'd add a switch somewhere on the phase level to create algorithms to go either one

Definite no-go, no need for that. Challenge organisers have already been warned that submissions can be a mix of exec and invoke containers. See https://grand-challenge.org/documentation/what-durations-are-available-for-algorithm-jobs/

[koopmant]

Still had a branch for this. If anyone would like to pick it up, please feel free.

[chrisvanrun]

FYI. The 'invoke' app has the potential to add quite some complexity to the example algorithm. I am not sure how to best go about this. Maybe we should have the 10% GC team go over a full example, and see where this stands?

Do we want to do this for the evaluation methods as well?

[jmsmkn]

I know, but users have been demanding this for years. It's really not that much more, see Thomas' branch.

Do we want to do this for the evaluation methods as well?

Yes, make it consistent.