check szip parameters for validity, plus all other outstanding szip work...

[edwardhartnett]

There are some invalid parameters for szip that HDF5 will only report at enddef time. This is not very nice for users! ;-)

So now setting szip with invalid parameters will be detected by netcdf-c, and the user will get a NC_EINVAL, and all will be right with the world again.

This PR includes all other outstanding szip PRs, which are required to get to this point. ;-)

Fixes #1615
Fixes #1622
Fixes #1473
Fixes #1618
Fixes #1616
Fixes #1612

[edwardhartnett]

@WardF this is ready to merge. It has passed all your CI and all my CI! ;-)

This will complete the szip work and close all szip related issues.

[edwardhartnett]

@WardF can we get this PR merged soon? It contains a lot of important changes and I don't want the PR to go stale. ;-)

Thanks,Ed

[DennisHeimbigner]

I concur with Ed, but for a different reason. The multifilter code
I built is going to mess strongly with Ed's szip-related changes.
So I need it be in the master as a stable base against which
to modify the multifilter code to match his changes.

[WardF]

Thanks, both of you!

[edwardhartnett]

Well @DennisHeimbigner that makes me nervous! ;-)

But I have changed the szip code to use nc_def_var_filter(), so hopefully that will make some of your changes easier.

Also @WardF thanks for all the help getting these changes merged. These new compression methods have already made a big difference for NOAA and all NOAA data users.

[DennisHeimbigner]

Adding support for multiple filters turned out to be a big change.
In order to avoid a lot of special casing, I chose to bring szip and deflate
into the more general framework.