Add Back2Local routing and benchmark analysis

[galtendorfer]

This PR adds Back2Local routing for MemPool and TeraPool, together with a
benchmark analysis and plotting pipeline built on the existing trace
infrastructure.

Changelog

Added

• Add Back2Local interconnect extensions and a tile-port rerouting mechanism for
  MemPool and TeraPool.
• Add benchmark data extraction and plotting for performance and communication
  analysis.
• Add a testbench observer for physical TCDM addresses after DAS scrambling,
  enabling correct destination analysis.

Fixed

• Fix DAS row-width and start-address handling in the DMA path.
• Fix DAS scrambling for partitions with offset start addresses.
• Fix the B matrix shape generated by generate_imatmul.

Validation

Tested a 256×64×256 integer matrix multiplication using the 4×4 DAS assembly
kernel on MemPool with Back2Local disabled and enabled, and on TeraPool with
Back2Local disabled, subgroup routing enabled, and subgroup plus group routing
enabled.

All five runs completed with 0 ERRORS out of 65536 CHECKS. Benchmark data
extraction and plot generation also completed successfully.

Checklist

• Automated tests pass
• Changelog updated
• Code style guideline is observed