feat[next][dace]: Support for sparse fields and reductions over lift expressions

[edopao]

Description

This PR adds support to DaCe backend for sparse fields and reductions over lift expressions.

The DaCe backend is configured with full-inlining of lifts in ITIR. An additional option is enabled, in combination with reduce-unroll, to fully inline lambda-like arguments. This option is required because the ITIR visitor in the DaCe backend is not able to translate reductions over lift expressions.

The inlined ITIR could contain the list_get operator. This PR adds translation capability for thelist_get operator.

In icon4py stencils, list_get is proceeded by the deref operator, in order to extract a 1D slice out of a multi-dimensional field. Baseline only supported deref operator on full-domain index, that returns a scalar value. When the deref indexing is partial, deref will return a 1D or ND slice from the original field, and the return type should be array. Therefore, this PR also adds support for deref with array return.

Requirements

• feat[next]: Add option to ITIR transformation to inline lambda args #1370

[philip-paul-mueller]

This is my first one. Further, I have one or two off topic questions, things that just pop up in my mind and would like to ask.

[edopao]

@philip-paul-mueller Thank you for the review, I have tried to address your comments. Please check again.

[philip-paul-mueller]

In the add_storage function we also create such variables, but we name them shape and stride.
We should use the same words in both functions. However, since they were part of #1379 I will change them there.