Partial Revert "Fix correlation on A100 CUTLASS GEMM Kernels (#127)"

[JRPan]

Summary

• Reverts the kernel issue (CTA scheduling) part of commit 987a23f
• Removes the pending_ctas queue mechanism from issue_block2core
• Removes allocated_ctas tracking from kernel_info_t
• Restores the original CTA scheduling behavior

This partial revert only affects the issue_block2core function and related data structures. The Perfect xbar interconnect changes from the original commit are not reverted.

Test plan

• Verify CTA scheduling works correctly with the reverted code
• Run correlation tests to ensure expected behavior

🤖 Generated with Claude Code

close accel-sim/accel-sim-framework#528
close accel-sim/accel-sim-framework#526

[JRPan]

Added commit to modernize latency array declarations with C++17 style:

• Replaced C-style arrays (unsigned int_latency[6], etc.) with std::array for type safety and bounds checking
• Added zero-initialization using {} for arrays to ensure all elements start at 0
• Added explicit = 0 initialization for scalar latency variables (sfu_latency, tensor_latency)
• Added #include <array> header