Add a DEAL_II_ASSUME macro.

[bangerth]

C++23 has a nice feature, [[assume(expr)]] that allows giving compilers hints about invariants that must be satisfied at a specific point -- see https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2022/p1774r8.pdf. I wanted to see whether we can perhaps use that in the "NO-DEBUG" definition of Assert to give the compiler a hint, but it turns out that I do not have access to a compiler that supports any of the various incompatible ways compilers have provided this sort of feature pre-C++23.

Let's see what the testers have to say. I would be curious to see what happens if anyone looks at the release mode library size without and with this patch, with a compiler that supports any of this.

(Separate question: We could of course also add the assumption after the if in the DEBUG version of the Assert macro. Opinions?)

[masterleinad]

(Separate question: We could of course also add the assumption after the if in the DEBUG version of the Assert macro. Opinions?)

In that case, the compiler might just compile the assert away, right?

[bangerth]

That's why I was only suggesting to do it in release mode where the assertion isn't there to begin with. But even in debug mode, I believe that such an assumption would only hold downstream, not upstream, or would it? If not, we'd place a call to a function the compiler can't look through (a barrier) between the assertion and the assume statement, to make sure the compiler can't know anything about upstream because it could have been changed in the opaque call.

[masterleinad]

In my opinion, it mostly makes sense in Release mode and I would rather not risk that some compiler would just discard the assert.

[masterleinad]

In terms of library size, I am seeing with clang:
new: 129246112
old: 129870368

[bangerth]

Oh, interesting! It's only half a percent, but if that translates to half a percent performance gain, then that ain't bad :-)

[bangerth]

This is now superseded by #16437.