sw: Add more complex Snitch application tests

[colluca]

Contributions:

• Bump Snitch cluster to include sequencer bug fix (included in the tapeout version)
• Bump PeakRDL-rawheader to include array element size bug fix
• Add vsim-run-batch-verify goal for apps with verification script (also in PD repo)
• Add AXPY, GEMM, FusedConcatLinear and MHA kernels to CI
• Add optimized 2D-GEMM kernel for Picobello
• Include first functions for a Picobello library
• Fix serial link preloading of Snitch binary. Add a generic write function that handles misaligned addresses, burst limits and 4KiB page crossings
• Add AXI dumper on memory tiles and analyzer script (@Lore0599 should we keep this here?)

[colluca]

As a general comment, we may want to prefix all functions as we do with snrt_ in the Snitch runtime, perhaps with pb_. The primary advantage to me being that it communicates clearly where they are defined.

[colluca]

A few more comments I forgot to add before.

[fischeti]

The snitch_cluster and common_cells bump look fine to me. I can remove the existing tapeout tag and recreate it once we merge this. I have some minor changes below.