CI: Check if par I/O is compatible with optimizations and parallelization (poly, vect, OMP, CSE, unrolling)

# Conflicts:
#	Compiler/src/exastencils/app/ir/IR_LayerHandler.scala
#	Compiler/src/exastencils/config/Knowledge.scala

devel/fix_cuda_nodePos_comm

See merge request exastencils/exastencils!72

Devel/matrix init fix

See merge request exastencils/exastencils!54

Fix for CUDA communication: Consider neighbor fragment accesses with "dity data" for memory transfers.

# Conflicts:
#   .gitlab-ci.yml

This reverts commit 613066bf

Refactor IR_AccessPattern to hold an IR_AccessFunction instance instead of a lambda. Refactor IR_IV_TemporaryBuffer and add member to distinguish fragment/block-wise storage.

Add read-after-write equality test for I/O of fields with transformed layout (stream-based I/O only).

Do not read/write databuffer as whole and contiguous data block when field layout trafo was performed.

This reverts commit 8abfadad