Drop JiT Guards in Most QF Source

[jeremylt]

Follow-up on #1534 and #1537, dropping QF source header guards where able

[jeremylt]

Fluids is the only one left - need to check for multiply included files and still guard those

Working on parallel changes in Ratel

[jrwrigh]

Is there any harm in adding #pragma once to all of them? It should work for CPUs and the GPU backends should handle the case fine, right?

[jeremylt]

#pragma once won't do anything for the GPU backends as all the included files are put into one long string buffer, but I think I have handled that with #1532. My plan is to first drop all guards that I can without using #pragma once and then see how many files that leaves. Then I'll see if #pragma once works with what's left.

[jeremylt]

Do you know if SYCL JiT handles #pragma once in a source file fine?

[jeremylt]

Turns out, ROCm doesn't like JiTing #pragma once so we can't do that.

[jrwrigh]

RIP, not sure about SYCL JIT, but I guess if ROCM doesn't then we probably shouldn't for any of the others.

Or (if we really wanted to) we could have CeedLoadSourceToInitializedBuffer_Private remove any #pragma once and rely on it to only allow one header for the JITed backends.

[jeremylt]

I'm hesitant to hide too much from the users. I think the way to go for now is to ask users to use #ifndef style guards under the umbrella of "mutually supported C99 + ROCm + CUDA" syntax. This PR drops 767 lines we don't need anymore, so I think we take the win

[jedbrown]

Is there any #pragma once in code that could be concatenated into a string, or is that always staying in headers that we just #include?

[jeremylt]

In this branch pragma once is not in any file that we turn into a string.

[jrwrigh]

I'm hesitant to hide too much from the users.

That's valid, but also we're already implementing what the #pragma once directive does. So it should be transparent to the user at all times, regardless. We basically just wrote a really simple preprocessor for your JIT string processing.

But it's 6-half-dozen-or-the-other to me.

[jeremylt]

The easy part is enabling the pragma for QFunctions. Backend JiT source is harder, but I have an idea

[jeremylt]

Ok, "turned on" #pragma once for QFunction source by stripping it out.

The way that Basis source is built means this won't work without another backend only interface, but I know what I have to expose for that to work.

[jrwrigh]

Few comments, overall looks good.