add htj2k option

[jcupitt]

to enable high-throughput j2k

see #6

[jcupitt]

Test with eg.:

$ time vips kakadusave k2.jpg x.jp2 --options Creversible=yes

real	0m0.091s
user	0m0.462s
sys	0m0.020s
$ time vips kakadusave k2.jpg x.jph --options Creversible=yes

real	0m0.080s
user	0m0.090s
sys	0m0.037s
$ ls -l x.jp2 x.jph
-rw-r--r-- 1 john john 2447603 Mar  5 18:29 x.jp2
-rw-r--r-- 1 john john 2681043 Mar  5 18:29 x.jph

You can see there's a dramatic drop in user time, and the file size is slightly larger.

The option is enabled automatically for filenames with a .jph suffix, or you can set the htj2k option if there's no filename available.

data = image.kakadusave_buffer(htj2k=True)

[scossu]

pyvips is still not detecting the extension:

>>> from pyvips import Image
>>> img = Image.new_from_file("x.jph")
>>> img.get("vips-loader")
'jp2kload'
>>> with open("x.jph", "rb") as fh:
...     img = Image.new_from_buffer(fh.read(), options="")
... 
>>> img.get("vips-loader")
'jp2kload_buffer'

Shouldn't the loader be kakaduload and kakaduload_buffer respectively?

[jcupitt]

The kakadu loader and saver don't take over libvips jpeg2000 handling, you have to invoke them explicitly with kakaduload and kakadusave.

[scossu]

OK, got it. My impression was that they would (as OPJ does, or doesn't it either?)

In either case it's not a big deal, I'll just add a fallback option.

[scossu]

Could you please look into the merge conflicts? I think I could resolve them but I'd rather rely on your judgment.

[jcupitt]

We could make them take over all jpeg2000 handling, but that should probably be another PR.

[scossu]

That would be ideal, thanks.

[jcupitt]

OK, I made #10