Fix multi-column filt for ndims(x)>2

[martinholters]

Our filt (and filt!) support multi-column input, working column-wise. That is extremely useful for e.g. stereo audio signals. The functions accept arbitrary-dimensional input, but only work properly for at most two dimensions.

Master:

julia> H = PolynomialRatio([0.1, 0.1], [1, -0.8]);

julia> filt(H, ones(5,3,2))
5×3×2 Array{Float64, 3}:
[:, :, 1] =
 0.1      0.1      0.1
 0.28     0.28     0.28
 0.424    0.424    0.424
 0.5392   0.5392   0.5392
 0.63136  0.63136  0.63136

[:, :, 2] =
 6.90139e-310  6.90139e-310  6.90139e-310
 6.90139e-310  6.90139e-310  6.90139e-310
 6.90139e-310  6.90139e-310  6.90139e-310
 6.90139e-310  6.90139e-310  6.90139e-310
 6.90139e-310  6.90139e-310  6.90139e-310

julia> filt(SecondOrderSections(H), ones(5,3,2))
5×3×2 Array{Float64, 3}:
[:, :, 1] =
 0.1      0.1      0.1
 0.28     0.28     0.28
 0.424    0.424    0.424
 0.5392   0.5392   0.5392
 0.63136  0.63136  0.63136

[:, :, 2] =
 0.1      0.1      0.1
 0.28     0.28     0.28
 0.424    0.424    0.424
 0.5392   0.5392   0.5392
 0.63136  0.63136  0.63136

While SecondOrderSections looks ok, running with -check-bounds=yes reveals that it's not quite sound either:

julia> filt(H, ones(5,3,2))
ERROR: BoundsError: attempt to access 5×3×2 Array{Float64, 3} at index [1, 1]
# [...]

julia> filt(SecondOrderSections(H), ones(5,3,2))
ERROR: BoundsError: attempt to access 5×3×2 Array{Float64, 3} at index [1, 1]
# [...]

I think once upon the time, Julia supported partial linear indexing, treating the last index in a getindex as a linear index for all remaining dimensions. Nowadays, using a CartesianIndex seems best for addressing the "column", so do that.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.90%. Comparing base (1edcf64) to head (3380f14).
Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #600   +/-   ##
=======================================
  Coverage   97.90%   97.90%           
=======================================
  Files          19       19           
  Lines        3252     3252           
=======================================
  Hits         3184     3184           
  Misses         68       68           

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

[wheeheee]

I suspect I removed that behaviour in a previous PR I made. But IIRC that required inbounds to work, so better this.