Generalize NA to take na predicate as argument (not just isnan)

[yha]

Useful for filtering SVector or GeometryTypes.Point arrays with NA(x->any(isnan,x)).
May be also useful to do NA(ismissing) in the future, but imfilter with missings doesn't work yet for other reasons.

I removed the line out[nanflag] .= nan(T), which may be breaking (changes the behavior of plain NA()). I'm not sure why it was there; seems to me like a bug.

[timholy]

Very nice!

[timholy]

Let's re-run this now on top of #174.

[codecov]

Codecov Report

Merging #172 into master will decrease coverage by 0.75%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #172      +/-   ##
==========================================
- Coverage   91.65%   90.90%   -0.76%     
==========================================
  Files           9        9              
  Lines        1222     1231       +9     
==========================================
- Hits         1120     1119       -1     
- Misses        102      112      +10     

Impacted Files	Coverage Δ
src/border.jl	94.90% <100.00%> (-0.61%)	⬇️
src/imfilter.jl	90.40% <100.00%> (-0.41%)	⬇️
src/ImageFiltering.jl	76.19% <0.00%> (-3.81%)	⬇️
src/mapwindow.jl	85.09% <0.00%> (-2.10%)	⬇️
src/borderarray.jl	89.70% <0.00%> (-1.34%)	⬇️
src/kernel.jl	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update aada1d2...d9c06e5. Read the comment docs.