Using CBLAS_INDEX and CBLAS_INT integer types in CBLAS

[weslleyspereira]

Description

Some of the collaborators think it is a good idea to have two indexes in CBLAS, let's say CBLAS_INDEX and CBLAS_INT (See the discussion in #461). In fact, this was the behavior in CBLAS before 4177968. Moreover, two indexes are still being used in MKL (CBLAS_INDEX and MKL_INT) and OpenBLAS (CBLAS_INDEX and blasint).

In this PR:

1. We move back to a cblas.h that uses two integer definitions, CBLAS_INDEX and CBLAS_INT. CBLAS_INDEX will only be used in the return of cblas_i*amax. With that, we restore an ABI that is compatible with other BLAS.
2. We choose the default value of CBLAS_INDEX to be size_t.

I am sure we have pros and cons regarding (2), and I hope we can discuss them in this thread.

Additional information:

• As in CBLAS before 83fc0b4, we expect the Fortran subroutine returns a signed integer of type CBLAS_INT. Therefore, cblas_i*amax must convert the return value to CBLAS_INDEX.
• OpenBLAS, MKL, GNU Scientific Library, and Numpy all use CBLAS_INDEX = size_t by default.
• The C interface for BLAS (https://www.netlib.org/blas/blast-forum/cinterface.pdf) indicates that, usually, CBLAS_INDEX = size_t.

Checklist

• The documentation has been updated
• If the PR solves a specific issue, it is set to be closed on merge.