Add hipSOLVER checklist tests (geqrf, orgqr/ungqr, potrs, syevj, gesvdj)

[Ozone-Whiskey]

Summary

Adds s/d/c/z tests for the hipSOLVER routine groups on the "most important routines to test" checklist in #411, in both the Fortran 2003 and Fortran 2008 binding tiers.

Routines covered (18 total, f2003 + f2008 each = 36 new test programs):

Group	Routines
QR factorization	Sgeqrf Dgeqrf Cgeqrf Zgeqrf
Generate Q	Sorgqr Dorgqr Cungqr Zungqr
Cholesky solve	Spotrs Dpotrs Cpotrs Zpotrs
Jacobi eigensolver	Ssyevj Dsyevj
Jacobi SVD	Sgesvdj Dgesvdj Cgesvdj Zgesvdj

Each routine is one commit (f2003 + f2008 pair + CMake registration).

Verification approach

Every test is self-verifying against a fixed small input, using a check robust to sign/phase/ordering ambiguity where relevant:

• geqrf: recover R from the output and confirm AH*A = RH*R.
• orgqr / ungqr: factorize with geqrf, generate Q, confirm Q**H*Q = I.
• potrs: pick a known x, form b = A*x for an SPD/HPD A, factorize with potrf, solve, confirm the solution recovers x.
• syevj: confirm Av = lambdav for each eigenpair.
• gesvdj: reconstruct A = USV**H and compare to the input.

All use the hipSOLVER workspace pattern (_bufferSize query + explicit dWork, create/destroy handle, and syevjInfo/gesvdjInfo params objects for the Jacobi routines).

Not included (blocked)

Two hipSOLVER checklist items could not be tested and are intentionally left out:

• ?getri — not present in the hipfort bindings at all (no hipsolver?getri interface is generated), so there is nothing to test.
• ?potrfBatched — the generated binding declares the batched pointer-array argument A as type(c_ptr) (by reference) rather than type(c_ptr), value. The C API is T* A[] (a T** passed by value), so every call faults at the next device sync with HIP error 700. Verified against the C header and confirmed that a corrected by-value interface works. This is a generator issue and should be fixed there; the test can be added once the binding is corrected.

Notes

• syevj/gesvdj have no generated array-pointer overload, so their f2008 tests pass device pointers via c_loc(...) to the generic interface; the other groups use the array-pointer form.
• All 36 new tests pass on gfx90a (MI210), ROCm 7.1.0, gfortran 13.3, both binding tiers. f2003 tests also compile with the Fortran 2008 pointer-interface macro OFF.

Part of #411.

Test plan

• Both tiers build and PASS for all 18 routines on gfx90a
• f2003 tests compile with the Fortran 2008 pointer-interface macro OFF

[amontoison]

Use .f08 extension for files in f2008/hipsolver.
Excellent work!
For the product with Q, Q', we could also check that ||C||_2 = ||op(Q) * C||_2.