Yes... I know LAPACK, but I found for BLAS 2 Level (Matrix-Vector multiplication) only this surboutine which performs the matrix-vector operation. LAPACK doesn't support matrices in CSR format as you can see here!
I do not understand what you are asking for and why, because source code and reference manual pages, while related, are quite different from each other as to the needs that they meet. The MKL routine that you mention is well-documented; therefore, if you have MKL and a compatible Fortran, C or other compiler, you have all that is needed to use that routine (and all routines in MKL) in your code.
MKL is commercial software. Intel generously provides free non-commercial-use binary licences for some versions of MKL. I am sure that Intel will not entertain requests for source code until you make persuasive arguments that binary releases will not work for you and that there is a benefit to Intel that accrues from releasing the source code to you. In the few cases where software companies issue source licenses for commercial software, the source code license may cost about two orders of magnitude more than a binary license, and will require that legal usage agreements be signed.
Having said that, I should point out that implementing a simple task such as M . v in any sparse matrix format is not a hard task. The algorithm is simple and well-conditioned, and writing this routine is a good way of learning about a specific sparse matrix format.