[Feature Request]: In IRM Model , allow for different learners for `g0` and `g1`

### Describe the feature you want to propose or implement

- The current interface is not flexible enough, e.g., to use pipeline based learners as for example in https://docs.doubleml.org/stable/examples/R_double_ml_pipeline.html

### Propose a possible solution or implementation

- The more complex the pipeline learners get, the less flexible the current API appears. Currently we only allow for 2 learners for `ml_g` and `ml_m` which should also be provided as the default in the future, but add an option to add separate learners etc. for `ml_g1` and `ml_g0` if required

### Did you consider alternatives to the proposed solution. If yes, please describe

_No response_

### Comments, context or references

_No response_

Begin with the IRM model learner interface and the linked R DoubleML pipeline example. Work out how separate learners for ml_g0 and ml_g1 should be represented while retaining the current ml_g and ml_m defaults, then validate the proposed API against pipeline-based learners.

Written by the indexing model from the issue text.