Refactor thread_dist

[alazzaro]

Currently, thread_dist is part the dbcsr_distribution. However, thread_dist is based on row_blk_sizes, which are part of dbcsr_type. So we can have a mix where a matrix has a thread distribution that doesn't match to its row_blk_sizes. This leads to an error when running the multiplications, where it can be that matrix_c has a different thread_dist than matrix_left. A workaround was put in place (see https://github.com/cp2k/dbcsr/issues/379 and https://github.com/cp2k/dbcsr/pull/404). Need to refactor to have thread_dist as part of dbcsr_type.