Don't log ignored errors unless debug logging is enabled

[bgilbert]

Avoid the dcm_error_newf() and dcm_error_free() if we won't log the message anyway.

With the current state of openslide/openslide#648, this avoids allocating, string-formatting, and freeing 690,658 dcm_filehandle_get_frame_number() errors when opening 3DHISTECH-2:

Before:
real	0m0.303s
user	0m0.224s
sys	0m0.077s

After:
real	0m0.210s
user	0m0.133s
sys	0m0.077s

[bgilbert]

Even with this fix I think there's an API problem. If the DICOM file is corrupt, causing dcm_filehandle_get_frame_number(NULL, ...) to fail, OpenSlide will mark the tile sparse and keep going. As a result, there's no opportunity to notice the problem later when that region is read. And if we explicitly check for DCM_ERROR_CODE_MISSING_FRAME, we're back to allocating and freeing hundreds of thousands of errors.

The MISSING_FRAME error could be a static one that doesn't get string-formatted or freed, but it feels like that gives libdicom too much of a magic understanding of how its API is being used. What I'd really like is

bool dcm_filehandle_has_frame(**error, *filehandle, column, row, *has_frame)

but that's an API change and probably adds some duplicated code.

What do you think?

[bgilbert]

I ended up adding an error check to openslide/openslide#648, so this PR is no longer directly relevant to OpenSlide (but still seems like a good idea).