[Bebop][RushB] Decouple runtime from Verilator

[daiyongyuan]

Summary

• base the refactor on the current topology-aware RushB implementation
• move command, event, wait, DMA request, queueing, timeout, and shutdown types into the shared RushB crate
• use core_id consistently; remove the obsolete accelerator/chip selection contract
• keep Verilator-specific RTL counters, DMA staging, and simulator stepping in the Verilator backend
• preserve mvin_mmio, Fence completion handling, per-Core routing, and single-threaded Verilator execution

This does not change the C++ FFI, RTL interface, batch regression configuration, or backend selection flow. BEMU keeps its synchronous per-Core backend because it does not expose Verilator-style Accepted/Completed events.

Verification

• cargo test -p bebop-rushb
• cargo clippy -p bebop-rushb --all-targets -- -D warnings
• cargo check -p bebop-verilator with generated Toy RushB RTL
• cargo check for the generated Toy BEMU crate
• git diff --check