Latent data race in WelchConfig

[palday]

The WelchConfig docstring has an interesting nugget:

DSP.jl/src/periodograms.jl

Lines 545 to 549 in d08699a

	!!! note
	WelchConfig precomputes an fft plan, and preallocates the necessary intermediate buffers.
	Thus, repeated calls to `welch_pgram` that use the same `WelchConfig` object
	will be more efficient than otherwise possible.

The latter part encourages re-use, but the former part hints that concurrent re-use may have a race condition. It might be useful to call this out explicitly so that a naive

config = WelchConfig(...)
psds = Vector{Periodogram}(undef, length(signals))
Threads.@threads for (idx, sig) in enumerate(signals)
    psds[idx] = welch_pgram(sig, config)
end

doesn't become a user footgun

[wheeheee]

Yeah, that's true for most of the configs. Maybe a warning is warranted but OTOH it should be obvious because it is stated that buffers are pre-allocated.

[wheeheee]

Then again, that isn't really implicit in the common understanding of config. I guess this should be done...