Integrating a slow licensed EDA backend: answer first, verify async

When adding Verdi (or any slow, licence-gated EDA tool) as a second engine behind an interactive feature like driver tracing, never put it on the critical path: answer each step from a fast local engine (e.g. slang static drivers + re-executing the active statement against waveform values), show it immediately, and fill in the tool's answer asynchronously with a per-step time budget, dropping it when the user moves on. Keep one warm batch session (verdi -nogui -play serving queries like srcActiveTrace) started lazily instead of per query, since startup is 15-70 s. The remaining bottleneck is usually waveform reads: run leaf reads concurrently, share in-flight reads, prioritize the fast engine's reads, and skip very wide signals for the secondary check.