[Math] Revert recent TMatrix optimizations

[guitargeek]

This PR reverts the following two commits:

• Rewrite the matrix multiplication function. rearrange the loops and us… #20365
• Remove openmp from MatrixMultiplication function #20639

I merged these PRs believing that their addressing a real problem for ATLAS users, as muhammadalhroob is an ATLAS collaboration member.

However, I learned later in another PR by the same contributor that the point of the changes to TMatrix was just to try out AI. That made me have second thoughts about the TMatrix optimizations, and I would suggest to revert them at this point for several reasons:

• TMatrix never claimed to have good performance, and users that need performant matrix multiplications use other libraries. So optimizing TMatrix with AI risks to break things without a clear benefit. If users are sticking with TMatrix, they seemingly don't mind slower matrix multiplications and appreciate instead the simplicity and stability of TMatrix.
• The new code introduced different code branches for different matrix sizes where we can't be sure if they are covered by the tests.
• We don't want to encourage AI-driven development for the fun of it if there is no motivation for the change.

I know that at least @silverweed is agreeing with the revert.

FYI, @muhammadalhroob. I hope you can understand the reasoning, and that this is nothing personal. We are ready to welcome and review your contributions to ROOT anytime! It was my totally mistake that I didn't ask about the motivation in your first PR.

[github-actions]

Test Results

    22 files      22 suites   3d 16h 54m 10s ⏱️
 3 791 tests  3 791 ✅ 0 💤 0 ❌
80 280 runs  80 280 ✅ 0 💤 0 ❌

Results for commit d7f7327.