Need to converge to a single firing rate function

[gabrevaya]

Currently, there are two functions for measuring mean firing rate: mean_firing_rate and firing_rate. @harisorgn, I think we should probably try to converge to a single function.

The function firing_rate has some serious performance issues, at least in the following case:

using Neuroblox
using StochasticDiffEq

@named fsi = Striatum_FSI_Adam(N_inhib = 10)
sys = get_system(fsi; simplify = true)
prob = SDEProblem(sys, [], (0.0, 10000.0), [])
sol = solve(prob, RKMil(); dt = 0.05, saveat = 0.05, abstol = 1e-2, reltol = 1e-2);
@btime firing_rate($fsi, $sol, threshold=-25)
# 125.693 s (1299235830 allocations: 78.25 GiB)

@btime let 
    spikes = detect_spikes($fsi, $sol; threshold=-25)
    t, fr_fsi = mean_firing_rate($spikes, $sol, trim_transient=500)
end
# 100.482 ms (841 allocations: 46.38 MiB)

I don't have time now to dig into what is causing that performance issue (I'll probably come back to this on Tuesday), but I guess it's due to:

    S = mapreduce(sparse_hcat, neurons) do neuron
        detect_spikes(neuron, sol; threshold, ts)
    end

in the detect_spikes function, similar to what used to happen in state_timeseries here: 87eb04b.

BTW, I've just added a dispatch for mean_firing_rate to account for EnsembleProblems and fixed the trimming of the initial transient in #455.

[harisorgn]

Agreed. I wrote firing_rate sloppily to debug my tutorial. The issue is most likely the multiple calls to detect_spikes. A solution with a single call is definitely possible, so more like what you have. We just need a dispatch where the first argument is a composite blox or a vector of bloxs.

[harisorgn]

Added more comments to #455 as I was going through your new dispatch. These comments are more relevant here though, apologies 😅 . I am linking so we don't forget.