Disable lossy coding of non-visual channels

[palemieux]

OpenJPH is not optimized for lossy coding of non-visual channels, e.g., depth maps.

This PR use heuristics to detect visual channels: visual channels are those whose suffix or full name is "r", "g", "b", "red", "green" or "blue". Even when requested globally by the caller, lossy coding is disabled on non-visual channels.

[peterhillman]

As discussed in the meeting, DWA has a similar approach to deciding which channels need to be lossless (internal_dwa_classifier.h)
It could be worth making the HTJ2K channel logic match DWA

[palemieux]

Looks like there is a bug deep in OpenJPH:

aous72/OpenJPH#310

[palemieux]

@cary-ilm What is the procedure if this patch depends on a version of OpenJPH where method signatures have changed?

[cary-ilm]

This PR goes into the lossy-htj2k branch, not main, and that branch will go into the 3.5 release, and we can make 3.5 depend on a modern OpenJPH.

It might be cleaner to submit a separate PR that bumps the OpenJPH version, in external as well as in the CI, merge that, then rebase this off of that.

[palemieux]

@cary-ilm the OpenJPH maintainer tells me he does not expect a tagged release (which the build system demands) until sometime in the future, so I guess this PR will remain until then.