Fix the variable transparency case that was broken by the #8255 commit.

[joa-quim]

The fix in the #8255 was actually very incomplete and left the working array still pointing to a zero filled array and hence this type of transparency was totally broken. This PR fixes that. But this PR also addressees other situations.

• For the variable transparency case, and given that it implies cresting a much bigger ps that considerably slows down the image rasterization, a new modifier +t is added to -Q to select true transparency. Otherwise, and by default, a blended transparency, that is an opaque combination of colors and alpha's, is computed.
• The single transparency value case was not working at all. This PR now does also the blended composition by default.
• The -Q docs are modified but I don't think they are telling the full truth. Debugging the code I don't see where a 2-values transparency layer is creating a true transparent figure. I always see the code path falling into the grdimage_img_set_transparency function that either copies a color or blends it. I don't know enough postscript to follow this matter any further.

If someone wants to play with a variable transparency file, here is one.

EDIT: The SC bellow sows a cut of the original of the transp_mix test. The top left image (that in fact is the second row, first col of the full image) shows the same grid pattern as the lower left. This grid pattern in PS is what we see when using the true variable transparency that works by drawing a rectangle for each pixel (that is why those PS are huge). So, at a given point the single value transparency (obtained for example with grdmix -A) must have been working, but as I said above, I have no idea how. I think we should replace that orig PS by the one produced now by the transp_mix.sh script and remove the GMT_KNOWN_FAILURE. @seisman Coul you do it after this RR is merged? I am not able to update original PS's.

[joa-quim]

Looks something is holding running the CI tests. I run them locally and no new failures were found.

[Esteban82]

I test the transp_mix.sh and I am getting some weird figures with linux.

If I use gmt begin transp_mix ps,png, I got this png (with this branch):

And I got this one on master:

But when I only ask for a png (gmt begin transp_mix png) I got this figure (in master and in this branch):

[joa-quim]

Strange. This is what I get with this branch after removing the GMT_KNOWN_FAILURE from transp_mix.sh and running ctest -R transp_mix

transp_mix.png

transp_mix_diff.png

[Esteban82]

Ok, I will try again at my home.

[Esteban82]

Ok, I will try again at my home.

I got this image. I don't know what is happening.

[seisman]

Sorry I don't get it. There are only documentation changes in this PR, no?

[joa-quim]

Ghrrr, I made the commit from the rst directory and the C files were left in my computer only. Now in #8613