DNMY: [FileFormats.MOF] change parse_as_nlpblock to false by default

[odow]

I've opened this PR, not because I want to merge it, but because I think it needs some discussion.

Currently, if you have a .mof.json file that contains nonlinear functions, they are parsed into an NLPBlock.

#2293 added support for parsing them as regular ScalarNonlinearFunctions, but made it opt-in behind a parse_as_nlpblock keyword.

To read in JuMP, you'd need to write:

using JuMP
m = JuMP.read_from_file("test.mof.json"; parse_as_nlpblock = false)

This PR proposes swapping the default, which is a breaking change if people expect the NLPBlock to exist, or if they're passing to a solver that doesn't support the new syntax yet.

We could potentially reduce the breakage by changing default_copy_to, so that, if a src model has ScalarNonlinearFunction but the dest model doesn't support them, we could convert to a NLPBlock. This would then only impact people using MOF.Model directly, not not some reading from a file in JuMP.

Or we could say that the nonlinear support in .mof.json was always experimental. It seems likely that very few people are actually using .mof.json files to store models, particularly nonlinear ones.

Or we could keep the default as-is and add a warning telling people to use the keyword argument, and that it might change the default in a future release.

Thoughts?

[odow]

Change to use_nlp_block to match NL files, and document this, and document it at the JuMP level.

[odow]

Discussed on call. Closing in favor of #2299