hw: Fix memory consistency between int and FP datapaths

[paulsc96]

ToDo:

• Handle CC parametric cases and non-FPU case (no CAQ needed)
• Basic debug in simulation
• Get all CI tests to run correctly (all except one pass: openmp_for_static_schedule fixed)
• Mind atomics: RMW "reads" must collide in CAQ!
• Add isochronous crossing to FPSS response channel (oops)
• Create minimal added test caq.c validating race fix
• Prevent multiple responses in FREP
• Add test for FREP fix
• Choose sensible defaults for parameters: need to avoid FREP performance issues -> Depth 8, Tag width 16
• Ensure failure of caq.c, caq_frep.c on prior system
• Verify internal exploration benchmarks
• Remove broken flashattention_2
• Fix Verilator tests in Docker container
• Resolve TODOs in RTL:
  • Are exceptions safe? (They should be)
  • Should response be default-cut for timing?
  • Can/should we respond on issue to save response RTT on collision?
• Evaluate area, timing synthesis and any degradations in implementation --> Future ToDo