Move convolution code to Convolutions.jl submodule

[wheeheee]

Not sure if xcorr belongs in there, and currently the deconv function calls filt, so that can't be moved into convolutions.jl as-is either.

[codecov]

Codecov Report

Attention: Patch coverage is 98.82353% with 3 lines in your changes missing coverage. Please review.

Project coverage is 98.13%. Comparing base (68341c4) to head (f54211c).

Files with missing lines	Patch %	Lines
src/convolutions.jl	98.81%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #613   +/-   ##
=======================================
  Coverage   98.13%   98.13%           
=======================================
  Files          19       20    +1     
  Lines        3277     3277           
=======================================
  Hits         3216     3216           
  Misses         61       61           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[martinholters]

Not that I'm necessarily opposed to this, but what's the benefit?

[wheeheee]

In #598 @dlfivefifty asked if convolutions could be moved out into a separate package. This PR would make it easier to do that, if the decision is made. One benefit would be that those who use only conv would have the choice to not load everything else here if they don't want to, and therefore be less affected by breaking releases in the rest of the package.

[dlfivefifty]

I wonder about whether FFTW.jl should be a hard dependency. Alternatively, a simpler package ConvolutionsBase.jl which only defines conv and conv! without implementing them might be a better solution.

For example, in InfiniteArrays.jl overloads of conv I don't actually need the fft at all:

https://github.com/JuliaArrays/InfiniteArrays.jl/blob/master/ext/InfiniteArraysDSPExt.jl

[martinholters]

Right, if we want to move the convolutions stuff to a seperate repo, there are a bunch of questions to answer. But I don't think this PR or #598 are the best places to discuss that. Maybe open a dedicated issue or discussion?

WRT this PR, I think we should wait until that discussion has settled before actually setting into anything in motion. But as a draft to try things out, it might indeed be helpful.

[dlfivefifty]

#615