Sync in changes from darktable. Fixes #158. Fixes #161. Fixes #162.

[LebedevRI]

All right, i guess it is time for a sync again.

Gist:

• Shitload of Nikon white-level fixes.
• Fixes for building with with libjpeg (as opposed to libjpeg-turbo)
• XSD for cameras.xml. Will allow us in darktable to do some better checks during CI. Would have prevented Remove stray ">" in PENTAX K-70 ID.
• nikon_override_auto_black hint is no more needed, and was removed.
• Heap-allocate big things that were stack-allocated before. Was part of cleanup to support running darktable on musl libc.
• Some more camera support, highlights:
  • Canon EOS 5D Mark IV, only full raw, i did not get m/s RAW samples yet.
  • Canon EOS 80D m/s RAW support, same as https://github.com/klauspost/rawspeed/pull/161
• A few white level fixes for Canon :(
• A bit of undefined behavior fixes
• Add missing Throw in RawDecoder::setMetaData() if no cam. Prevents loading of broken images when guessing (failOnUnknown) is disabled.

Now what this does not fix: These 3 cameras got disabled:

Camera ["NIKON CORPORATION", "NIKON D3X", "12bit-compressed"] has too high white level: 4096 (>= ((2^12)-1), which is 4095)
Camera ["NIKON CORPORATION", "NIKON D3X", "12bit-uncompressed"] has too high white level: 4096 (>= ((2^12)-1), which is 4095)
Camera ["NIKON", "E8400", "12bit-compressed"] has too high white level: 4095 (>= ((2^12)-1), which is 4095)
Camera ["NIKON", "E8400", "12bit-uncompressed"] has too high white level: 4095 (>= ((2^12)-1), which is 4095)
Camera ["NIKON", "E8800", "12bit-compressed"] has too high white level: 4095 (>= ((2^12)-1), which is 4095)
Camera ["NIKON", "E8800", "12bit-uncompressed"] has too high white level: 4095 (>= ((2^12)-1), which is 4095)

They can be trivially re-enabled after a raw sample for each is found, to check white levels. Not sure, should i enable them in this pr, for upstreaming? May complicate upstreaming later on.

And, it was found that NIKON D5100 is lying (duh), in this case it marks compressed raws as uncompressed, https://redmine.darktable.org/issues/11268, or at least it seems so.

Special thanks to @schenlap.

[LebedevRI]

@klauspost friendly ping :)

[LebedevRI]

And some more fixes:

• Fix for name mangling when reading DNG's from nikon. Now that nikon cameras in cameras.xml have only entries with mode="???", it did not work, thus breaking name demangling.
• Proper(?, at least better) order for black levels in Rw2 raws. It seems to be in CFA order, and the original code did not work for LX15 highest ISO - the sep black levels were in different order compared to DNG.

[LebedevRI]

@axxel i guess now you know how to contribute :) I have simply fixed it the way it did fix it. I did not benchmark and did not know about https://github.com/klauspost/rawspeed/commit/8e0ad84d91bce504a22ef6933de885f687bcf7c4

Benchmark it, and based on the results, possibly change code and PR it. As you can see, upstreaming changes into rawspeed is .. time-y process, so you may want to PR to https://github.com/darktable-org/darktable :)

[LebedevRI]

@klauspost thank you!