Investigate introducing a robust resource consumption operation

The protocol currently relies on the `clk` operation to estimate the prover's resource consumption to execute a given program. However, this is a poor metric, as operations vary drastically in the amount of memory or processor time they require to execute (e.g. `noop` takes orders of magnitude less time than `hperm`).

We should instead investigate introducing an operation that returns the number of "trace cell contributions" at this point in the program execution. This probably means that we're adding a system column to track the number of trace cells computed so far (which this new operation reads from), and increment it by the number of trace cell contributions depending on the operation (e.g. `emit` would increment by `1 * num_core_columns`, whereas `hperm` would increment by `32 * num_hasher_columns + 1 * num_core_columns`).