[Monitor] Reactor to use `bi`'s `proto_trace.rs` interface to avoid memory blowup on long waveforms

On the `ethmac` benchmark, the monitor runs out of memory after ~10 mins (even on our servers).
The `bi` executable also eventually runs out of memory on the same waveform, but after ~1 hour.
(Aside: we will investigate other Wishbone benchmarks with shorter waveforms for the time being)

From reading the `bi` code, it looks like `bi` keeps track of tail pointers when accumulating the protocol trace, so "cloning" the scheduler just means updating the tail pointer instead of copying the entire Vec of protocol entries, which seems much more efficient.
On the other hand, the current `monitor` implementation naively clones the entire `Scheduler` struct every time it explores a different execution path, which entails copying the entire `Scheduler.output_buffer` `Vec`, which becomes expensive when there are more paths.

Ideally, we would refactor `monitor` to use the same mechanism as `bi` (by using the data structures in `bi/src/proto_trace.rs`) to avoid this blow-up in memory.