fix: L3 max bandwidth buffer size for more accurate results

[harp-intel]

This pull request updates the logic for measuring L3 cache performance in the benchmarking scripts. The main goal is to more accurately configure buffer sizes for different L3 cache tests, distinguishing between idle latency and maximum bandwidth scenarios.

Benchmark script improvements:

• Added a new buffer setup script, mlcBufferSetupL3BW, to calculate the buffer size for L3 max bandwidth benchmarks based on 80% of the per-thread L3 cache share, ensuring a minimum buffer size proportional to L2 cache per thread.
• Updated the L3MaxBandwidthBenchmarkScriptName benchmark to use the new mlcBufferSetupL3BW buffer setup and removed the -u flag from its command-line options.

Comment and script clarity:

• Clarified comments to distinguish between L3 idle latency (single-threaded, 4x L2 size) and L3 max bandwidth (multi-threaded, 80% per-thread L3 share) buffer setup logic.