Reject distributed matrices in DiagoLapack

[Growl1234]

DiagoLapack is a local LAPACK eigensolver and requires replicated Hamiltonian and overlap matrices. However, some MPI execution paths could pass a distributed local matrix block to it through either diag() or diag_pool().

For example, with 8 MPI processes, each process may own a 52 × 26 block of a global 104 × 104 matrix. DiagoLapack previously passed the global dimension, N = 104, to LAPACK while providing only the local block. LAPACK then accessed memory beyond the allocated buffers, causing heap corruption such as:

free(): invalid size
munmap_chunk(): invalid pointer
double free or corruption

This PR:

• verifies in both diag() and diag_pool() that the Hamiltonian and overlap matrices are replicated nlocal × nlocal matrices before calling LAPACK;
• reports a clear error when distributed local blocks are passed and instructs users to select ScaLAPACK or ELPA;
• verifies that the output eigenvector buffer has sufficient dimensions;
• allocates the full RWORK and IWORK arrays before the LAPACK workspace query;
• handles the xSYGVX/xHEGVX INFO value according to the LAPACK specification.

For distributed matrices, users are instructed to select ScaLAPACK or ELPA instead. Single-process LAPACK calculations remain work as before.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[ZhouXY-PKU]

Nice job!