Deprecate eltype-specific convert methods

[timholy]

These are technically pirating and cause quite a few invalidations. Now that we have generic broadcasting they are no longer necessary. xref JuliaGraphics/ColorTypes.jl#196.

This first submission of the PR does not fix depwarns in the tests, because I want to see if everything passes before doing so.

Also xref #122. We should merge these two at around the same time, along with any other deprecations, to reduce the number of times we annoy users.

[codecov]

Codecov Report

Merging #131 into master will increase coverage by 0.45%.
The diff coverage is 88.88%.

@@            Coverage Diff             @@
##           master     #131      +/-   ##
==========================================
+ Coverage   60.50%   60.95%   +0.45%     
==========================================
  Files           9        9              
  Lines         400      397       -3     
==========================================
  Hits          242      242              
+ Misses        158      155       -3     

Impacted Files	Coverage Δ
src/convert_reinterpret.jl	62.06% <ø> (-1.35%)	⬇️
src/deprecations.jl	47.05% <88.88%> (+47.05%)	⬆️

---

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 3beb222...f984124. Read the comment docs.

[dehann]

+1 for this fix after coming across the same issue.

[timholy]

I declare open season on making more disruptive changes.

Here's my plan:

• I will get the rest of the Images ecosystem (or as much as I can discover) locally to the point where it's not triggering depwarns in internal code.
• We release new patch-level versions of all the packages
• We consider whether we need to signal to the users on Julia 1.5 and above. We can force depwarns to be printed like this:

const depwarnidx = (i = 1; T = Base.JLOptions; while i <= fieldcount(T) fieldname(T, i) === :depwarn && break; global i += 1; end; fieldoffset(T, i) + 1)
function forced_depwarn(msg, sym)
    oldval = unsafe_load(cglobal(:jl_options, UInt8), depwarnidx)
    unsafe_store!(cglobal(:jl_options, UInt8), Int8(true), depwarnidx)
    ret = try
        Base.depwarn(msg, sym)
    finally
        unsafe_store!(cglobal(:jl_options, UInt8), oldval, depwarnidx)
    end
    return ret
end

Personally I think we need to do that, given that most users probably say using Images and don't explicitly control the version of ImageCore they use.

[dehann]

Great thanks for fixing this! Patch level updates sound like a good idea to me. Forcing depwarn printouts (or at least some kind of default=true) sounds like the right thing again. Type piracy can be tricky, so any guidance won't go amiss.

I'm a major fan of Images.jl btw, and likely to build a lot of application code using Images in the future.

[timholy]

Great to hear! Don't hesitate to advertise your work to us!