Raising: unroll invariant-yield do-whiles

[wsmoses]

A rotated strided-copy loop (k = tid; do { copy(k); } while (k < n) with the increment folded by the range analysis) arrives as an scf.while whose after region yields only loop-invariant values or pass-throughs of the condition-forwarded ones: every iteration past the second would repeat the second's state exactly, so a third implies the original program never terminates. Unroll it to body(init); if (cond) body(invariant), selecting the forwarded results accordingly.

This is one of the normalizations behind the bilininteg_curlcurl_pa family raising strict in the MFEM CUDA→xla-gpu campaign (#2968). Stacked on #3006 (base pb/buffer-normalizations) for the preprocessing-round scaffolding.

Includes a lit test of the rotated strided-copy shape raising end to end.

🤖 Generated with Claude Code

https://claude.ai/code/session_016zErYp7upmqr4NHfhod9UD

[wsmoses]

Superseded by #3071, which moves this unroll into canonicalize-parallel (so it runs before the final llvm-to-affine-access and frees allocas carried through the loop, tmop assemble diag) and fixes a soundness hole: this PR accepts any pass-through of a forwarded condition operand as invariant, so its own @dowhile test (%next = %k + 16, scf.condition(%cond) %next, scf.yield %j) is a real strided loop that gets unrolled to two iterations. #3071 requires the forwarded value itself to be invariant and the exit test to be pure over the carried state.