The goal is to correct the documentation generated for the CCM CCGR
fields. Based on how the combiner works, these fields are generally
considered combine-able across all chips. So when we generate code for
the 1061 CCM, we're permitted to use the fields from the 1011 CCM. The
1011 CCM has many more reserved gates than the 1062 CCM, so those end up
leaking into the documentation.

This commit drops the fields that are marked "reserved" in their
human-readable description. It's not clear why these are provided in the
SVD. By dropping these reserved fields, we change the "field plurality"
heuristic that the combiner uses to understand equivalent fieldsets.
Notice how the CCGR documentation corrects itself in the 1061 CCM. We
also generate less code, and less chance for humans to poke at reserved
registers.

I'm not a fan of hard-coding this behavior into the SVD importer. But it
was easy to prototype. I might consider a transform that does this
across all document-able items.

I didn't use a 'contains("reserved")' predicate for the field skip.
There are some fields that are documented with "all other values are
reserved," so that larger-reaching predicate would incorrectly remove
those fields. I briefly scanned these changes to make sure removals
seemed appropriate.

Supply a regex that matches an IR path. If the combiner matches that
path, it always signals "not equivalent." This prevents the combiner
from combining the element.

Apply the configuration to CCM CCGR registers. This fixes the
documentation that would not be fixed by the previous commit (it's a
superficial documentation change, but nevertheless a good test that the
"never combine" behavior is working).

Similar to strict enum names, this configuration instructs the combiner
to compare enum variant descriptions, the human-readable strings
associated with items. If they're not equal, then the enums cannot be
combined.

When applied to IOMUXC peripherals, this ensures that the documentation
associated with each enum variant is correct. The next commit checks in
the updates.

Briefly spot-checked the 1061 IOMUXC to make sure the diff of some
registers are correct. See the parent commit for context.

The 1020, 1050, and 1060 SVDs qualify these fields as reserved, so the
previous commits scrubbed the fields. Patch the SVDs to re-introduce the
field.