[hlsl-out] More matCx2 fixes

[teoxoy]

fixes #1837

[jimblandy]

This doesn't handle the case of arrays of matrices; see my comments on the code.

We should probably remove the reference to #1837 in src/back/hlsl/mod.rs.

[teoxoy]

Fixed matCx2s nested inside global arrays however this issue is also present for structs containing matCx2 arrays.

[teoxoy]

@jimblandy this should now be good to go. I think we have all our matCx2 bases covered now 😅.

I added some more comments - I hope they are helpful.

[jimblandy]

I need to look at this again tomorrow. Comments thus far, nothing too serious:

It might be good to add __mat, __get, and __set to the list of reserved prefixes passed to Namer::reset in `hlsl::Writer::reset.

It might be worthwhile going though help.rs and similar code to look for other prefixes we've neglected to tell Namer to avoid. So far I've found ret_ (for the typedefs generated by write_wrapped_constructor_function) and Construct.

Or, maybe just start all generated names with __, and then only warn Namer about that?

[teoxoy]

Regarding the Namer, I'd say we go for

maybe just start all generated names with __, and then only warn Namer about that?

since WGSL already doesn't allow identifiers that start with two underscores.

[jimblandy]

Okay - having thought on this a little more, I'm inclined to merge this as-is for the moment.

But the current design (which predates this PR) feels focused on identifying special cases. In the longer term, we need to look hard for some useful generalizations, that will make it easier to see that all the cases are being covered.